Intelligent control method and system for noninvasive ventilator

Abstract

The invention discloses an intelligent control method and system for a noninvasive respirator, and relates to the technical field of intelligent control of respirators. According to the method, airway flow and pulse wave signals are obtained through signal collection and respiratory phase recognition, the end-expiratory pause time period is positioned, a time-frequency image is constructed and preprocessed to obtain a high-resolution enhanced time-frequency energy image, and cardiac oscillation components are recognized and separated to obtain pure respiratory flow signals. The autonomous inspiration effort is judged through double triggering conditions, and air supply control is achieved; verifying parameters are evaluated and adaptively updated based on the respiratory cycle, and personalized configuration parameters are locked; the problems of false triggering and leakage triggering air supply caused by cardiac airway flow oscillation interference generated by cardiac pulsation in the ventilation process of the noninvasive breathing machine are effectively solved, and the accuracy and reliability of air supply triggering of the breathing machine are remarkably improved.

Description

A smart control method and system for a non-invasive ventilator Technical Field

[0001] This invention relates to the field of intelligent control technology for ventilators, specifically to an intelligent control method and system for a non-invasive ventilator. Background Technology

[0002] Non-invasive ventilators, with mask-based non-invasive ventilation at their core, rely on flow or pressure triggering mechanisms to sense the patient's spontaneous inspiratory effort and achieve synchronized air delivery. They are widely used for ventilatory support in conditions such as respiratory failure and sleep apnea. In clinical applications, the patient's heartbeat is transmitted to the airway through the chest cavity, forming periodic and regular cardiogenic airway flow oscillations. These oscillation signals are similar in characteristics to spontaneous inspiratory signals, easily interfering with trigger judgment. Current mainstream non-invasive ventilation control mostly uses fixed threshold triggering and baseline leak compensation, which cannot effectively distinguish between cardiogenic and spontaneous inspiratory oscillations. Oscillating oscillations and true inspiratory signals are crucial. When the oscillation amplitude reaches the trigger threshold, it can trigger false triggers without voluntary inspiratory effort, leading to hyperventilation, respiratory rhythm disturbances, and patient-ventilator asynchrony. Reducing sensitivity to avoid false triggers can mask weak inspiratory signals, making inspiratory effort undetectable and resulting in missed triggers, further exacerbating respiratory muscle work and insufficient ventilation. These triggering abnormalities caused by cardiac interference are more prevalent in emaciated individuals, those with increased cardiac output, and those with low airway resistance. Adjusting the trigger threshold alone is insufficient to balance sensitivity and interference resistance. Current technologies lack the ability to identify, separate, and suppress cardiac oscillations in real time, making it difficult to fundamentally resolve the contradiction between false triggers and missed triggers, thus limiting the synchronization and safety of non-invasive ventilation. Therefore, an intelligent control method that can accurately distinguish between respiratory and cardiac signals is urgently needed to improve ventilation stability and patient tolerance.

[0003] To address the aforementioned shortcomings, a technical solution is provided. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of false triggering and missed gas delivery caused by cardiac airway flow oscillation interference generated by heart pulsation during non-invasive ventilation, and to propose an intelligent control method for non-invasive ventilators.

[0005] The objective of this invention can be achieved through the following technical solution: An intelligent control method for a non-invasive ventilator, comprising: S1, signal acquisition and respiratory phase recognition: acquiring airway flow signals in the airway tubing of the non-invasive ventilator through a flow sensor, extracting real-time heart rate by combining the fingertip pulse wave signal acquired by a pulse oximeter, performing respiratory phase analysis on the airway flow signals, and identifying the end-expiratory apnea period in the current respiratory cycle; S2, time-frequency image construction and preprocessing: extracting the airway flow signals during the end-expiratory apnea period as the analysis window signal, generating a time-frequency energy matrix through continuous wavelet transform and mapping it into a two-dimensional time-frequency energy image, and performing image preprocessing on the two-dimensional time-frequency energy image; S3, image feature recognition. Separation: The region of interest for the frequency band of cardiogenic oscillations is cropped from the preprocessed image. Cardiogenic oscillations are identified through morphological opening, binarization, and connected component screening. A soft mask is generated and weighted to suppress oscillation energy. The pure respiratory flow signal, free of interference, is reconstructed through inverse wavelet transform. S4, Trigger Judgment and Ventilation Control: Spontaneous inspiration is triggered by a dual trigger condition on the pure flow signal. When the condition is met, a ventilation command is generated and the ventilator ventilation state is switched. S5, Adaptive Parameter Update and Verification: Based on the trigger judgment results of multiple consecutive respiratory cycles, the oscillation separation parameters in the process of cardiogenic oscillation image feature recognition and separation are adaptively updated, and the update effect is verified and evaluated.

[0006] Furthermore, the specific operation steps of S1 are as follows: the airway flow signal is collected by the flow sensor of the non-invasive ventilator airway at a preset sampling frequency, and the fingertip pulse wave signal is collected by the pulse oximeter of the patient's fingertip. After amplification, filtering and noise reduction, the real-time heart rate value is extracted; the original airway flow signal is filtered by a bandpass filter to filter out invalid components and retain the effective respiratory and cardiogenic oscillation related components; the preprocessed airway flow signal is traversed with a fixed-length sliding window and the root mean square value is calculated. When the root mean square value is continuously lower than the preset resting flow threshold and the duration is longer than the preset pause duration threshold, it is determined to be an end-expiratory pause period, and the start and end times of the end-expiratory pause period are recorded.

[0007] Furthermore, the specific operation steps of S2 are as follows: Based on the identified end-expiratory pause period, the corresponding airway flow signal is extracted as the analysis window signal; the analysis window time length is obtained based on the cardiac cycle acquisition multiple, the cardiogenic oscillation fundamental frequency, and the minimum window length; Morlet wavelet is selected as the mother wavelet, and continuous wavelet transform is performed on the analysis window signal to obtain wavelet transform coefficients; based on the wavelet transform coefficients, the time-frequency energy value is obtained and a time-frequency energy matrix is ​​constructed; the time-frequency energy matrix is ​​normalized and mapped into a two-dimensional time-frequency energy image; median filtering, background subtraction, and contrast stretching enhancement are sequentially performed on the image; the scale parameter spacing is reduced within a preset frequency range centered on the cardiogenic oscillation fundamental frequency; bilinear interpolation is performed on the corresponding frequency row region to densify it; and the corresponding frequency row region in the contrast-stretched image is replaced to form a locally high-resolution enhanced two-dimensional time-frequency energy image along the frequency axis.

[0008] Furthermore, the specific operation steps of S3 include: obtaining the upper and lower boundaries of the expected frequency band based on the fundamental frequency of the cardiogenic oscillation and the preset half-width of the frequency band; cropping all image rows within the upper and lower boundaries of the expected frequency band from the enhanced two-dimensional time-frequency energy image to form a frequency band region of interest image; performing a horizontal morphological opening operation on the frequency band region of interest image; converting the opening operation result image into a binary image based on a binarization threshold; performing eight-neighbor connected component labeling; eliminating invalid connected components based on pixel area and the aspect ratio of the minimum bounding rectangle; performing time axis column projection and summing on the retained connected components to obtain a one-dimensional time-energy projection curve; detecting the peak value through normalized autocorrelation operation and obtaining the relative deviation by combining it with the heart rate cycle; determining the cardiogenic oscillation based on the relative deviation and calculating the confidence index of the cardiogenic oscillation.

[0009] Furthermore, the specific operation steps of S3 also include: when it is determined that there is cardiogenic oscillation, mapping the pixel coordinates of each connected component in the frequency band region of interest image after screening back to the global coordinate system of the enhanced two-dimensional time-frequency energy image, setting edge transition bands along the frequency axis and time axis outside the boundary of the connected component respectively and constructing cosine gradual decay weights, setting different regional mask values ​​based on the weights to form a cardiogenic oscillation soft mask; performing weighted suppression processing on the time-frequency energy based on the cardiogenic oscillation soft mask and the confidence index to reduce the energy proportion of the oscillation region, performing inverse wavelet transform on the suppressed image based on the mother wavelet function, reconstructing the two-dimensional time-frequency energy data into a one-dimensional time domain signal, and obtaining a pure respiratory flow signal that has eliminated the interference of cardiogenic oscillation.

[0010] Furthermore, the specific operation steps of S4 are as follows: A fixed-step sliding window is used to traverse the pure respiratory flow signal after removing cardiogenic oscillations. Within each window, the flow rate increase trend and volume accumulation trend are calculated simultaneously. Dual trigger conditions for flow rate increase rate and volume accumulation are set. Only when both trigger conditions are simultaneously met and the duration exceeds a preset confirmation time is it determined that the patient has spontaneous inspiratory effort and a delivery trigger command is generated, controlling the ventilator to switch from expiratory to positive inspiratory pressure ventilation. If only the flow rate increase rate condition is met, it is marked as a suspected trigger state, and the sliding window length is extended by 1.5 times for continued monitoring. If the dual trigger conditions are still not met after extending the window monitoring, the suspected trigger mark is removed. If neither trigger condition is met, it is determined that there is currently no spontaneous inspiratory effort, and the ventilator maintains the current positive end-expiratory pressure ventilation state. After each trigger determination or delivery trigger, the multidimensional feature label is synchronously stored in the trigger event log.

[0011] Furthermore, the specific operation steps of S5 are as follows: An evaluation cycle is defined as multiple consecutive respiratory cycles. Within the evaluation cycle, the total number of triggered events, the number of suppressed false triggers, and the number of missed triggers confirmed by the chest and abdominal motion sensor are counted. The number of suppressed false triggers and the number of missed triggers confirmed by the chest and abdominal motion sensor are compared with the total number of consecutive respiratory cycles within the evaluation cycle to obtain the false trigger suppression rate and the missed trigger rate. A comprehensive performance index is obtained by combining the false trigger rate benchmark value. Based on the comparison results between the comprehensive performance index and the preset threshold, parameters are adaptively adjusted. If the separation of the cardiogenic oscillation signal is too strong, the frequency band half-width is reduced and the binarization threshold coefficient is increased; if the separation is insufficient, the frequency band half-width and the binarization threshold coefficient are increased. After parameter adjustment, the parameters are recalculated in the next evaluation cycle. If the comprehensive performance index is greater than the preset threshold for three consecutive evaluation cycles, the current oscillation separation parameters are locked and stored in the ventilator storage unit as a patient-specific ventilation configuration file for subsequent continuous ventilation control.

[0012] The second aspect of this invention provides an intelligent control system for a non-invasive ventilator, comprising: a signal acquisition and respiratory phase recognition module: integrating flow and pulse oximetry sensors to acquire airway flow and fingertip pulse wave signals, analyzing and locating the end-expiratory apnea period through a sliding window to complete respiratory phase recognition; a time-frequency image construction and preprocessing module: extracting the analysis window signal based on the end-expiratory apnea period, performing continuous wavelet transform to generate a two-dimensional time-frequency energy image, and performing image preprocessing on the two-dimensional time-frequency energy image; an image feature recognition and separation module: cropping out the region of interest in the frequency band, identifying cardiogenic oscillations through morphological processing and connected component screening, generating a soft mask, suppressing oscillation components through weighted summation, and reconstructing and outputting a pure respiratory flow signal based on inverse wavelet transform; a trigger judgment and ventilation control module: traversing the pure flow signal, calculating flow and volume trends, determining spontaneous inspiratory effort through dual conditions, generating a ventilation command and switching the ventilation state, and synchronously recording trigger event tags; and an adaptive parameter update and verification module: calculating comprehensive performance indicators based on respiratory cycle statistical trigger parameters and adaptively adjusting oscillation separation parameters, locking and storing the parameters as a personalized ventilation configuration file for the patient after they stabilize.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention achieves accurate identification of respiratory phase by collecting airway flow and pulse wave signals, constructs and optimizes time-frequency images to identify cardiogenic oscillations, effectively separates oscillations from respiratory flow components through soft mask weighted suppression, determines spontaneous inspiratory effort and controls air delivery based on dual triggering conditions, evaluates performance through respiratory cycles and adaptively updates verification parameters to lock in personalized ventilation configurations; it significantly improves the accuracy of non-invasive ventilator air delivery triggering, effectively suppresses false triggering problems caused by cardiogenic oscillations, reduces the probability of missed triggers, improves the level of intelligent control of ventilators and the safety and adaptability of clinical ventilation, and provides patients with non-invasive ventilation support that better meets their physiological needs. Attached Figure Description

[0014] Figure 1 is a flowchart of the method of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example: As shown in Figure 1, an intelligent control method for a non-invasive ventilator includes signal acquisition and respiratory phase recognition, time-frequency image construction and preprocessing, image feature recognition and separation, trigger judgment and gas delivery control, and adaptive parameter update and verification.

[0017] S1. Signal Acquisition and Respiratory Phase Recognition: Airway flow signals are continuously acquired at a preset sampling frequency using a flow sensor installed in the non-invasive ventilator's airway tubing. Simultaneously, a pulse oximeter is worn on the patient's fingertip, and the fingertip pulse wave signal is acquired in real time using photoplethysmography. After amplification, filtering, and noise reduction of the pulse wave signal, the time interval between adjacent pulse wave peaks is detected to obtain the real-time heart rate value. The acquired raw airway flow signal is preprocessed using a bandpass filter with a passband range of 0.1Hz to 25Hz to remove DC offset components, power frequency interference components, and high-frequency electromagnetic noise components, retaining effective respiratory and cardiogenic oscillation-related components, resulting in a preprocessed airway flow signal. The cardiogenic oscillation-related components include the cardiogenic oscillation signal component and the cardiogenic oscillation time-frequency energy component. A fixed-length sliding window is used to traverse the preprocessed airway flow signal along the time axis, and statistical analysis is performed within each sliding window using a formula. The root mean square value of the airway flow signal is calculated, where, This indicates the number of sampling points within the sliding window. Indicates the first [number]th ... The airway flow rate values ​​at each sampling point; when the root mean square value is continuously lower than the preset resting flow rate threshold and the duration reaches the preset pause duration threshold, the current period is determined to be the end-expiratory pause period, and the start and end times of the end-expiratory pause period are recorded to complete the respiratory phase recognition and end-expiratory pause period location.

[0018] S2. Time-frequency image construction and preprocessing: Based on the end-expiratory apnea period determined in S1, the airway flow signal segment from the start time within the end-expiratory apnea period is extracted as the analysis window signal. The formula for calculating the analysis window length is as follows: ;in, Indicates the length of the analysis window. Represents the maximum value function. Indicates the heart rate collection multiple. Indicates the fundamental frequency of cardiac oscillations. The minimum window length refers to the minimum window duration for time-frequency analysis. The Morlet wavelet is selected as the mother wavelet function, and a continuous wavelet transform is performed on the airway flow signal within the analysis window to obtain the wavelet transform coefficients. Based on these coefficients, the formula is used to... The time-frequency energy value is obtained through calculation, where, Represents the continuous wavelet transform coefficients. Let represent the scale parameter and time parameter in the core parameters of the wavelet transform coefficients, respectively. Based on the time-frequency energy values ​​at each scale and time position, a time-frequency energy matrix is ​​formed, where the row direction corresponds to the scale parameter of the wavelet transform coefficient, and the matrix row number... Each column corresponds one-to-one with the discrete scale index; the column direction corresponds to the time parameter of the wavelet transform coefficient, and the matrix column number. Each location corresponds one-to-one with a discrete-time position number; through the formula The time-frequency energy matrix is ​​normalized and mapped to a two-dimensional time-frequency energy image with a grayscale value range of [0, 255]. In a two-dimensional time-frequency energy image, the first... line, number The grayscale value of the column pixels, Indicates the first The discrete scale, the first Time-frequency energy values ​​at discrete time locations This represents the global maximum time-frequency energy value in the time-frequency energy matrix. This represents the rounding function; the window for executing the function on the generated two-dimensional time-frequency energy image is... The median filtering operation for pixels involves applying a formula to each row of pixels along the time axis of the filtered image. Calculation yields the first Linear background energy estimate, Represents the total number of pixels along the time axis; expressed by the formula. The calculation yields the second-dimensional time-frequency energy image after background subtraction, where the first... line, number The grayscale value of each pixel; if the grayscale value of the 2D time-frequency energy image is negative after background subtraction, it is set to 0; contrast stretching enhancement is performed on the 2D time-frequency energy image after background subtraction: the 2nd percentile of all non-zero grayscale values ​​in the 2D time-frequency energy image after background subtraction is calculated. and the 98th percentile The grayscale values ​​are remapped to the range [0, 255] using the following formula: ;in, This indicates that after contrast stretching and enhancement, the second time-frequency energy image shows the... line, number The grayscale values ​​of the column pixels; centered on the cardiogenic oscillation fundamental frequency, in Within the range, the original scale parameter spacing is reduced to half of the original spacing. ,in, This indicates the preset encryption frequency radius. Indicates the encryption ratio; locates the original two-dimensional time-frequency energy image. The corresponding frequency row region; bilinear interpolation is performed on the original wavelet coefficient matrix of the located frequency row region to generate a refined time-frequency image row with encryption; the corresponding frequency row region in the contrast-stretched image is replaced with the refined time-frequency image row with the encrypted time-frequency image row to form a local high-resolution enhanced two-dimensional time-frequency energy image along the frequency axis.

[0019] S3. Image Feature Recognition and Separation: Based on the cardiogenic oscillation fundamental frequency and combined with the preset frequency band half-width, the upper and lower boundaries of the expected frequency band are calculated using the following formula: , ;in, These correspond to the upper and lower boundaries of the expected frequency band, respectively. This represents the preset frequency band half-width; from the enhanced two-dimensional time-frequency energy image, all image rows with frequency ranges within the upper and lower boundaries of the expected frequency band are cropped to form the region of interest image. ,in, This represents a local row index within the region of interest (ROI) in the frequency band image, corresponding only to the frequency range where cardiogenic oscillations occur; a horizontal morphological opening operation is performed on the ROI image: a horizontal linear structuring element is constructed using the formula... The length of the horizontal linear structuring element is calculated, where, This indicates the sampling frequency of the signal in the analysis window. Indicates the heartbeat cycle, The preset structuring element length ratio coefficient is used. Based on the horizontal linear structuring element, the region of interest (ROI) image in the frequency band is subjected to erosion followed by dilation to obtain the opening operation result image. The erosion operation is used to eliminate fine, non-horizontal noise textures; the dilation operation is used to restore the complete shape of the horizontal stripes; the global grayscale mean of the opening operation result image is extracted. and global grayscale standard deviation ; through formula The binarization threshold is calculated, where, The binarization threshold coefficient is used; pixels with gray values ​​greater than the binarization threshold in the opening operation result image are marked as foreground, and the remaining pixels are marked as background, generating a binarized image; the eight-neighbor connected component labeling is performed on the binarized image, and the key parameters of each connected component are counted, including pixel area and minimum bounding rectangle aspect ratio, and connected components that do not meet the elimination rules are filtered out. The elimination rules are as follows: (1) Area less than The connected components, where, This represents the minimum area threshold for a connected component. This represents the lower limit coefficient for the area ratio. (1) The number of rows in the region of interest image of the frequency band; (2) Connected components whose aspect ratio of the minimum bounding matrix is ​​less than a preset threshold; Perform column projection summation on the retained connected components in the time axis direction to obtain a one-dimensional time-energy projection curve; Perform normalized autocorrelation operation on the one-dimensional time-energy projection curve, and use the formula The normalized autocorrelation function is obtained; where, The one-dimensional time-energy projection curve represents the first... The values ​​that a column can take. Represents the pixel offset along the time axis; based on the normalized autocorrelation function, detects pixels other than... The first local maximum peak outside of the specified range; the pixel offset corresponding to the local maximum peak is marked as... ; Pixel offset The actual time interval is converted and its relative deviation from the calculated heart rate cycle is determined using the following formula: ;in, This indicates the relative deviation; if the relative deviation is less than the preset matching deviation threshold... The presence of cardiogenic oscillation was determined using the formula. The confidence index is calculated, where, This represents the normalized autocorrelation function at pixel offset. The function value at the location; if the relative deviation is greater than the preset matching deviation threshold, it is determined that there is no cardiogenic oscillation, and subsequent mask generation is skipped, directly outputting the original airway flow signal; when it is determined that there is cardiogenic oscillation, a cardiogenic oscillation soft mask image is generated based on the preserved connected components: the pixel coordinates of each connected component in the frequency band region of interest image after screening are mapped back to the global coordinate system of the enhanced two-dimensional time-frequency energy image; outside the boundary of the mapped connected components, it is extended along the frequency axis and time axis respectively. Pixels serve as edge transition zones, where: , ;in, These are the frequency and time-based transition band scaling factors, respectively; within the edge transition band, a cosine-gradient attenuation weight is constructed, as shown in the following formula: ;in, This represents the attenuation weight value of pixels within the edge transition zone. This represents the pixel distance from the current pixel to the boundary of the connected component. This indicates the width of the edge transition band corresponding to the direction of the current pixel. Representing pi, used to control the period of the cosine function; setting the mask value of pixels inside the connected component to 1, and the mask value of pixels in the transition zone to 1. The mask value of pixels outside the transition zone is set to 0; if a pixel is simultaneously located in the overlapping area of ​​the frequency and time direction transition zones, the product of the frequency and time direction attenuation weight values ​​is taken as the final mask value, forming a cardiogenic oscillation soft mask with the same size as the two-dimensional time-frequency energy image.

[0020] The time-frequency energy is suppressed using a weighted method based on a soft mask of cardiogenic oscillations and a confidence index. The weighted suppression formula is as follows: ;in, This represents the suppressed enhanced two-dimensional time-frequency energy image of the first... line, number The energy value corresponding to the column pixel. The first two-dimensional time-frequency energy image before suppression represents the first... line, number The energy value corresponding to the column pixel. The value represents the soft mask value for cardiac oscillations. For the enhanced two-dimensional time-frequency energy image after complete suppression processing, based on the mother wavelet function, the two-dimensional time-frequency energy data is reconstructed into a one-dimensional time-domain signal through wavelet inverse transform to obtain a pure respiratory flow signal after removing cardiac oscillation interference, thereby achieving effective separation of the cardiac oscillation component and the respiratory flow component in the airway flow signal.

[0021] S4. Trigger Judgment and Gas Delivery Control: A fixed-step sliding window is used to traverse the pure respiratory flow signal after removing cardiac oscillation interference. Within each sliding window, the flow rate upward trend and volume accumulation trend are calculated simultaneously. The specific calculation formula is as follows: Flow Rate Upward Trend: Accumulated capacity value: ;in, This indicates the total number of sampling points contained within the current sliding window. These represent the positions within the sliding window, respectively. Time and the The corresponding pure respiratory rate value at any given moment. This indicates the original sampling interval of the signal; a dual trigger condition joint determination mechanism is set: the first trigger condition is the flow rate rise condition: The second triggering condition is the capacity accumulation condition: ;in, This indicates the preset threshold for determining the rate of increase in traffic. This indicates the preset volume accumulation threshold. Only when both the first and second trigger conditions are met simultaneously, and the duration of this condition exceeds the preset confirmation duration, is it determined that the patient is currently exhibiting spontaneous inspiratory effort. A ventilator delivery trigger command is immediately generated, controlling the ventilator to switch from expiratory to positive inspiratory pressure (POP) ventilation. If only the first trigger condition is met, but the second trigger condition is not, the current state is marked as a suspected trigger state, and the current sliding window length is extended to 1.5 times the original window length for continued monitoring. If both trigger conditions are not met after extending the window monitoring, the suspected trigger mark is removed, and no effective spontaneous inspiratory effort is determined. If neither the first nor the second trigger condition is met, no spontaneous inspiratory effort is directly determined, and the ventilator maintains the current positive end-expiratory pressure ventilation state. After each trigger determination or delivery trigger, the current pure respiratory flow amplitude, original airway flow amplitude, confidence index for cardiac oscillation identification, and the percentage of pixels in the cardiac oscillation soft mask are used as multidimensional feature labels and simultaneously stored in the trigger event log.

[0022] S5. Adaptive Parameter Update and Validation: An evaluation cycle is defined as multiple consecutive respiratory cycles. Key parameters are statistically analyzed within each cycle, including the total number of trigger events, the number of suppressed false triggers, and the number of missed triggers confirmed by the chest and abdominal motion sensor. The false trigger suppression rate is calculated based on the ratio of the number of suppressed false triggers to the total number of consecutive respiratory cycles within the evaluation cycle. The missed trigger rate is calculated based on the ratio of the number of missed triggers confirmed by the chest and abdominal motion sensor to the total number of consecutive respiratory cycles within the evaluation cycle. ; through formula The comprehensive performance index is obtained through calculation, among which, This represents the baseline spurious trigger inhibition rate when the cardiac oscillation isolation function is not enabled. This represents the false trigger suppression rate after enabling the cardiac oscillation separation function. Parameters are adaptively adjusted based on the comprehensive performance index and a preset threshold: when the comprehensive performance index is greater than the preset threshold, the current cardiac oscillation separation and trigger determination parameters are deemed valid and remain unchanged; when the comprehensive performance index is less than the preset threshold and the missed trigger rate is greater than the preset maximum allowable missed trigger rate, it is determined that the cardiac oscillation signal separation is too strong, resulting in excessive suppression of ventilation triggering, and a rollback adjustment is performed: reducing the frequency band half-width and increasing the binarization threshold coefficient to narrow the range of the cardiac oscillation soft mask; when the comprehensive performance index is less than the preset threshold and the false trigger suppression rate after enabling the cardiac oscillation separation function is greater than the preset maximum allowable false trigger suppression rate, it is determined that the cardiac oscillation signal separation is insufficient and the false trigger suppression is inadequate, and an enhancement adjustment is performed: increasing the frequency band half-width and decreasing the binarization threshold coefficient to expand the range of the cardiac oscillation soft mask; after the parameter adjustment is completed, statistics and calculations are re-performed in the next evaluation cycle. If the comprehensive performance index is greater than the preset threshold for three consecutive evaluation cycles, the current oscillation separation parameters are locked, including the frequency band half-width, binarization threshold coefficient, structural element length ratio coefficient, and matching deviation threshold. The oscillation separation parameters are used as the current patient's personalized ventilation profile and stored in the ventilator storage unit for subsequent continuous ventilation control.

[0023] An intelligent control system for a non-invasive ventilator includes: a signal acquisition and respiratory phase recognition module: integrating a flow sensor and a pulse oximeter, continuously acquiring airway flow signals and fingertip pulse wave signals, performing amplification, filtering, and noise reduction preprocessing on the signals, calculating the root mean square value of airway flow through a sliding window, determining the end-expiratory apnea period, and completing respiratory phase recognition and localization; a time-frequency image construction and preprocessing module: extracting and analyzing window signals based on the identified end-expiratory apnea period, generating a time-frequency energy matrix through continuous wavelet transform and mapping it to a grayscale image, sequentially performing median filtering, background subtraction, and contrast stretching enhancement, while locally densifying the frequency axes surrounding the cardiogenic oscillation fundamental frequency, outputting a high-resolution enhanced two-dimensional time-frequency energy image; and an image feature recognition and separation module: cropping the frequency band region of interest of the enhanced two-dimensional time-frequency energy image, and performing morphological opening operations, etc. The system employs a multi-level marketing and connected component screening method to identify cardiogenic oscillations. It calculates a confidence index and generates a soft mask, then uses weighted suppression to remove oscillatory components from the time-frequency energy. Based on wavelet inverse transform, it reconstructs and outputs a pure respiratory flow signal. The trigger judgment and ventilation control module iterates through the pure respiratory flow signal using a sliding window, calculating the flow rate increase trend and volume accumulation trend. It jointly determines spontaneous inspiratory effort using dual trigger conditions, generating a ventilation trigger command when the conditions are met to control the ventilator's switching ventilation state. Simultaneously, it records multi-dimensional feature labels of the trigger events. The adaptive parameter update and verification module calculates the false trigger suppression rate, missed trigger rate, and comprehensive performance index based on key trigger parameters statistically analyzed during the respiratory cycle. Based on the comparison between the comprehensive performance index and preset thresholds, it adaptively adjusts the oscillation separation parameters. Once the parameters stabilize, they are locked and stored as a personalized ventilation configuration file for the patient, continuously optimizing the control effect.

[0024] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent control method for a non-invasive ventilator, characterized in that, include: S1. Signal Acquisition and Respiratory Phase Recognition: Airway flow signals within the non-invasive ventilator's airway are acquired using a flow sensor. Real-time heart rate is extracted by combining this with fingertip pulse wave signals from a pulse oximeter. Respiratory phase analysis is performed on the airway flow signals to identify the end-expiratory apnea period within the current respiratory cycle. S2. Time-Frequency Image Construction and Preprocessing: The airway flow signal during the end-expiratory apnea period is extracted as the analysis window signal. A time-frequency energy matrix is ​​generated through continuous wavelet transform and mapped to a two-dimensional time-frequency energy image. Image preprocessing is then performed on the two-dimensional time-frequency energy image. S3. Image Feature Recognition and Separation: Cardiogenic oscillations are cropped from the preprocessed image. The frequency band region of interest is used to identify cardiogenic oscillations through morphological opening, binarization, and connected component screening. A soft mask is generated and weighted to suppress oscillation energy. The pure respiratory flow signal, free of interference, is reconstructed through inverse wavelet transform. S4, Trigger judgment and ventilation control: The pure flow signal is jointly judged by dual trigger conditions to trigger spontaneous inspiration. When the conditions are met, a ventilation command is generated and the ventilator ventilation state is switched. S5, Adaptive parameter update and verification: Based on the trigger judgment results of multiple consecutive respiratory cycles, the oscillation separation parameters in the process of cardiogenic oscillation image feature recognition and separation are adaptively updated, and the update effect is verified and evaluated.

2. The intelligent control method for a non-invasive ventilator according to claim 1, characterized in that, The specific operation steps of S1 are as follows: the airway flow signal is collected by the flow sensor of the non-invasive ventilator airway at a preset sampling frequency, and the fingertip pulse wave signal is collected by the pulse oxygen sensor of the patient's fingertip. After amplification, filtering and noise reduction, the real-time heart rate value is extracted. The raw airway flow signal was filtered using a bandpass filter to remove invalid components and retain the effective respiratory and cardiogenic oscillation-related components. The preprocessed airway flow signal is traversed using a fixed-length sliding window, and the root mean square value is calculated. When the root mean square value is continuously lower than the preset resting flow threshold and the duration is longer than the preset pause duration threshold, it is determined to be an end-expiratory pause period, and the start and end times of the end-expiratory pause period are recorded.

3. The intelligent control method for a non-invasive ventilator according to claim 1, characterized in that, The specific operation steps of S2 are as follows: Based on the identified end-expiratory pause period, the corresponding airway flow signal is extracted as the analysis window signal; the analysis window time length is obtained based on the cardiac cycle acquisition multiple, the cardiogenic oscillation fundamental frequency, and the minimum window length; Morlet wavelet is selected as the mother wavelet, and continuous wavelet transform is performed on the analysis window signal to obtain wavelet transform coefficients; the time-frequency energy value is obtained based on the wavelet transform coefficients and a time-frequency energy matrix is ​​constructed; the time-frequency energy matrix is ​​normalized and mapped into a two-dimensional time-frequency energy image; median filtering, background subtraction, and contrast stretching enhancement are performed on the image in sequence. Centered on the cardiogenic oscillation fundamental frequency, the scale parameter spacing is reduced within a preset frequency range. Bilinear interpolation is then performed on the corresponding frequency row region to encrypt it. This process replaces the corresponding frequency row region in the contrast-stretched image, forming a locally high-resolution enhanced two-dimensional time-frequency energy image along the frequency axis.

4. The intelligent control method for a non-invasive ventilator according to claim 1, characterized in that, The specific operation steps of S3 include: obtaining the upper and lower boundaries of the expected frequency band based on the fundamental frequency of the cardiogenic oscillation and the preset half-width of the frequency band; cropping all image rows within the upper and lower boundaries of the expected frequency band from the enhanced two-dimensional time-frequency energy image to form a region of interest image of the frequency band; performing a horizontal morphological opening operation on the region of interest image of the frequency band; converting the opening operation result image into a binary image based on a binarization threshold; performing eight-neighbor connected component labeling; eliminating invalid connected components based on pixel area and the aspect ratio of the minimum bounding rectangle; performing time axis column projection and summing on the retained connected components to obtain a one-dimensional time-energy projection curve; detecting the peak value through normalized autocorrelation operation and obtaining the relative deviation by combining it with the heart rate cycle; determining the cardiogenic oscillation based on the relative deviation and calculating the confidence index of the cardiogenic oscillation.

5. The intelligent control method for a non-invasive ventilator according to claim 4, characterized in that, The specific operation steps of S3 further include: when it is determined that there is cardiogenic oscillation, mapping the pixel coordinates of each connected domain in the frequency band region of interest image after screening back to the global coordinate system of the enhanced two-dimensional time-frequency energy image, setting edge transition bands along the frequency axis and time axis outside the boundary of the connected domain and constructing cosine gradual attenuation weights, setting different regional mask values ​​based on the weights to form a cardiogenic oscillation soft mask; performing weighted suppression processing on the time-frequency energy based on the cardiogenic oscillation soft mask and the confidence index to reduce the energy proportion of the oscillation region, performing inverse wavelet transform on the suppressed image based on the mother wavelet function, reconstructing the two-dimensional time-frequency energy data into a one-dimensional time domain signal, and obtaining a pure respiratory flow signal that has eliminated the interference of cardiogenic oscillation.

6. The intelligent control method for a non-invasive ventilator according to claim 1, characterized in that, The specific operation steps of S4 are as follows: use a fixed step size sliding window to traverse the pure respiratory flow signal after removing cardiogenic oscillations, calculate the flow rate rise trend and volume accumulation trend simultaneously in each window, set dual trigger conditions for flow rate rise rate and volume accumulation, and determine that the patient has spontaneous inspiratory effort and generate a delivery trigger command only when the dual trigger conditions are met at the same time and the duration exceeds the preset confirmation duration, and control the ventilator to switch from expiration to positive inspiratory pressure delivery state. If only the flow rate rise condition is met, it is marked as a suspected trigger state, and the sliding window length is extended by 1.5 times to continue monitoring. If the dual trigger conditions are still not met after extending the window monitoring, the suspected trigger mark is removed. If neither of the dual trigger conditions is met, it is determined that there is currently no spontaneous inspiratory effort, and the ventilator maintains the current positive end-expiratory pressure ventilation state. After each trigger determination or air delivery trigger, the multidimensional feature label is synchronously stored in the trigger event log.

7. The intelligent control method for a non-invasive ventilator according to claim 1, characterized in that, The specific operation steps of S5 are as follows: taking multiple consecutive respiratory cycles as an evaluation cycle, and counting the total number of triggering events, the number of suppressed false triggers, and the number of missed triggers confirmed by the chest and abdominal motion sensor within the evaluation cycle. The number of suppressed false triggers and the number of missed triggers confirmed by the chest and abdominal motion sensors are compared with the total number of consecutive respiratory cycles within the evaluation period to obtain the false trigger suppression rate and the missed trigger rate. Combined with the false trigger rate benchmark, a comprehensive performance index is obtained. Based on the comparison results of the comprehensive performance index and the preset threshold, the parameters are adaptively adjusted. If the separation of the cardiogenic oscillation signal is too strong, the frequency band half-width is reduced and the binarization threshold coefficient is increased. If the separation is insufficient, the frequency band half-width is increased and the binarization threshold coefficient is increased. After parameter adjustment, the parameters are recalculated in the next evaluation period. If the comprehensive performance index is greater than the preset threshold for three consecutive evaluation periods, the current oscillation separation parameters are locked and stored in the ventilator storage unit as the patient's personalized ventilation configuration file for subsequent continuous ventilation control.

8. A system applied to the intelligent control method of a non-invasive ventilator according to any one of claims 1-7, comprising: Signal acquisition and respiratory phase recognition module: integrates flow and pulse oximetry sensors to acquire airway flow and fingertip pulse wave signals, and uses a sliding window to analyze and locate the end-expiratory apnea period to complete respiratory phase recognition; Time-frequency image construction and preprocessing module: extracts the analysis window signal based on the end-expiratory apnea period, performs continuous wavelet transform to generate a two-dimensional time-frequency energy image, and performs image preprocessing on the two-dimensional time-frequency energy image; Image feature recognition and separation module: crop out the region of interest in the frequency band, identify cardiogenic oscillations through morphological processing and connected component screening, generate a soft mask, suppress oscillation components through weighted summation, and reconstruct the output pure respiratory flow signal based on wavelet inverse transform; Trigger judgment and ventilation control module: traverse the pure flow signal, calculate flow and volume trends, determine spontaneous inspiratory effort through dual conditions, generate ventilation commands and switch ventilation states, and synchronously record trigger event tags; Adaptive parameter update and verification module: Based on respiratory cycle statistical trigger parameters, it calculates comprehensive performance indicators and adaptively adjusts oscillation separation parameters. After the parameters stabilize, they are locked and stored as a personalized ventilation configuration file for the patient.

Images