A pediatric chest wall patch type wireless intelligent respiratory sound monitoring system and method

Abstract

The application discloses a kind of paediatrics chest wall pasting type wireless intelligent breath sound monitoring system and method, it is related to medical monitoring technical field.The paediatrics chest wall pasting type wireless intelligent breath sound monitoring system, including chest wall pasting type acquisition terminal, abnormal gate control module and hierarchical transmission module.Chest wall pasting type acquisition terminal gathers the chest wall breath sound signal of sick child and extracts breath sound feature vector and attachment quality feature;Abnormal gate control module executes the double baseline abnormal gate control algorithm of attachment disturbance correction, generates normal monitoring mark or abnormal trigger mark;Hierarchical transmission module switches low-power inspection mode and high-fidelity event upload mode accordingly.The application separates by disposable pasting ring piece and host body, attachment quality feature correction and double baseline abnormal gate control linkage hierarchical transmission, reduce the influence of attachment disturbance on abnormal judgment, consider sick child wearing, host multiplexing, low-power monitoring and abnormal audio reservation.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, specifically to a pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system and method. Background Technology

[0002] In the diagnosis and treatment of pediatric respiratory diseases, auscultation of breath sounds is an important means of observing the condition and evaluating the treatment effect. During the progression of diseases such as pneumonia, bronchiolitis, and acute asthma exacerbations, characteristic changes such as moist rales, wheezing, decreased breath sounds, or bilateral asymmetry of breath sounds are often observed. These characteristic changes will dynamically change as the disease progresses, improves, or fluctuates.

[0003] With the development of miniature sensors, wireless communication, and intelligent recognition technologies, respiratory sound monitoring devices have gradually evolved from traditional handheld stethoscopes to electronic stethoscopes, wearable data acquisition devices, and patch-type monitoring devices, providing a foundation for continuous and objective recording, transmission, and analysis of respiratory sounds. However, pediatric patients exhibit significant differences in chest size, activity level, cooperation, and baseline respiratory sounds. Furthermore, pediatric ward, outpatient screening, and home follow-up scenarios place higher demands on wearing comfort, skin contact component replacement, low-power operation, abnormal event retention, and subsequent data utilization. Existing general-purpose respiratory sound monitoring devices still suffer from insufficient compatibility when directly applied to pediatric scenarios.

[0004] The limitations of existing technologies include at least the following problems: Current pediatric breath sound monitoring largely relies on handheld auscultators, vest-style chest strap devices, or integrated patch devices. Handheld auscultators require manual, timed operation, and the collection time point is easily misaligned with the window of rapid changes in the patient's condition. Vest-style chest strap devices fix the collection components by covering or binding them, which can easily lead to insufficient wearing compliance and auscultation point displacement due to the feeling of restraint, skin pressure, the child's pulling, and changes in body position, making it difficult to stably correspond changes in breath sound characteristics to changes in the patient's condition. Integrated patches integrate the sensor, circuitry, battery, and skin contact layer, making it difficult to independently replace the skin contact part when reused. When used once, electronic components are discarded along with consumables, making it difficult to balance infection control, host reuse, and usage costs. During continuous monitoring, high-fidelity real-time uploading increases power consumption, while intermittent acquisition may miss short-term abnormal respiratory sound events. Existing abnormality judgments mostly rely on fixed thresholds or single audio classification, which leads to instability when dealing with differences in the child's age, position, basic respiratory sounds, and attachment status. Furthermore, the abnormality judgment results are not linked to low-power inspection and high-fidelity uploading, and there is a lack of a structured processing mechanism for screening, desensitizing, and associating real abnormal respiratory sound segments. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system and method, which solves the problems of unstable acquisition terminal, host reuse and hospital infection control, low-power continuous monitoring, reliability of anomaly detection, and difficulty in coordinating the use of real abnormal audio structures in existing pediatric respiratory sound monitoring.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system, comprising: a chest wall patch-type acquisition terminal, including a main body and a disposable patch ring; the main body is provided with a pickup window, a respiratory sound acquisition module, a signal preprocessing module, a feature extraction module, and a wireless communication module; the disposable patch ring is detachably connected to the main body and has a hollow area corresponding to the pickup window; the respiratory sound acquisition module acquires the respiratory sound signal of the child's chest wall through the hollow area; the signal preprocessing module processes the respiratory sound signal of the child's chest wall into a digital respiratory sound signal; the feature extraction module extracts respiratory sound feature vectors and patch quality features from the digital respiratory sound signal; and an anomaly gating module is used to perform a dual-baseline anomaly gating algorithm for patch perturbation correction on the respiratory sound feature vectors and patch quality features, including: based on the respiratory sound features within the initial acquisition window... The system generates an individualized respiratory sound baseline template, an attachment stability benchmark, and respiratory sound baseline parameters for the same age group based on the amount of data, attachment quality characteristics, the child's age group, and body position information. It then generates an attachment perturbation coefficient based on the attachment quality characteristics within the subsequent monitoring time window. This coefficient is used to correct the respiratory sound feature vector within the subsequent monitoring time window. Finally, it generates a routine monitoring marker or anomaly trigger marker based on the deviation of the corrected respiratory sound feature vector from the individualized respiratory sound baseline template and the respiratory sound baseline parameters for the same age group. A graded transmission module controls the chest wall patch-type acquisition terminal to enter a low-power inspection mode and upload compressed respiratory sound segments based on the routine monitoring marker. Based on the anomaly trigger marker, it controls the chest wall patch-type acquisition terminal to enter a high-fidelity event upload mode and upload the original respiratory sound audio stream. After the anomaly trigger marker fades, it controls the chest wall patch-type acquisition terminal to return to the low-power inspection mode.

[0007] Furthermore, the attachment and acquisition structure of the chest wall patch-type acquisition terminal is set as follows: the disposable patch is set as a hollow ring structure, and the hollow area corresponds coaxially with the sound pickup window on the bottom surface of the main body; a medical pressure-sensitive adhesive layer is set on the bottom surface of the disposable patch, a medical polyurethane film substrate is set in the middle of the disposable patch, and a detachable connection structure that cooperates with the main body is set on the top surface of the disposable patch; the main body is fixed on the disposable patch through the detachable connection structure, so that the sound pickup window corresponds to the hollow area of ​​the disposable patch; after the chest wall patch-type acquisition terminal is attached to the chest wall of the child through the disposable patch, the hollow area forms an air coupling cavity between the chest wall of the child and the sound pickup window; the respiratory sound acquisition module receives the respiratory sound signal of the child's chest wall through the air coupling cavity and transmits the respiratory sound signal of the child's chest wall to the signal preprocessing module.

[0008] Further, the specific steps of the feature extraction module to extract the respiratory sound feature vector and attachment quality features are as follows: The digital respiratory sound signal is segmented according to the monitoring time window to obtain a windowed digital respiratory sound signal; respiratory frequency, energy envelope, spectral centroid, zero-crossing rate, and Mel frequency cepstral coefficients are extracted from the windowed digital respiratory sound signal to form a respiratory sound feature vector; low-frequency baseline drift, instantaneous energy mutation, spectral centroid jump variable, and effective energy ratio of the signal are extracted from the windowed digital respiratory sound signal to form attachment quality features; the respiratory sound feature vector and attachment quality features are bound according to the same monitoring time window to obtain respiratory sound window feature data with attachment status markers; the respiratory sound window feature data with attachment status markers is sent to the anomaly gating module.

[0009] Further, the specific steps for generating the attachment perturbation coefficient and correcting the respiratory sound feature vector within the subsequent monitoring time window are as follows: Obtain the attachment quality characteristics within the initial acquisition window, and calculate the stable distribution of low-frequency baseline drift, instantaneous energy mutation, spectral centroid variable, and the proportion of effective signal energy within the initial acquisition window to generate an attachment stability benchmark; obtain the attachment quality characteristics within the subsequent monitoring time window, and calculate the offset of the attachment quality characteristics relative to the attachment stability benchmark within the subsequent monitoring time window; generate the attachment perturbation coefficient based on the offset amplitude. The attachment perturbation coefficient increases with the increase of low-frequency baseline drift, instantaneous energy mutation, and spectral centroid variable, and increases with the decrease of the proportion of effective signal energy; use the attachment perturbation coefficient to compensate for the energy envelope and spectral centroid in the respiratory sound feature vector within the subsequent monitoring time window to obtain the corrected respiratory sound feature vector; input the corrected respiratory sound feature vector into the individualized respiratory sound baseline template and the deviation calculation process of the respiratory sound baseline parameters of the same age group.

[0010] Further, the specific steps for generating routine monitoring markers or abnormal trigger markers are as follows: The individual baseline deviation is obtained based on the Mahalanobis distance between the corrected breath sound feature vector and the individualized breath sound baseline template; the group baseline deviation is obtained based on the Mahalanobis distance between the corrected breath sound feature vector and the breath sound baseline parameters of the same age group; the individual and group baseline deviations are input into a dual baseline deviation fusion function to obtain an abnormality gating value; the abnormality gating value within the continuous monitoring time window is accumulated in chronological order to obtain the abnormality gating duration; an abnormality trigger marker is generated when the abnormality gating duration reaches the abnormality triggering condition, and a routine monitoring marker is generated when the abnormality gating duration does not reach the abnormality triggering condition.

[0011] Furthermore, the specific transmission control steps of the hierarchical transmission module are as follows: After receiving the regular monitoring marker, the chest wall patch acquisition terminal is controlled to acquire digital respiratory sound signals according to the inspection acquisition cycle; the digital respiratory sound signals within the inspection acquisition cycle are compressed and encoded to obtain compressed respiratory sound segments; the compressed respiratory sound segments are uploaded through the wireless communication module, and the chest wall patch acquisition terminal is controlled to enter the sleep state after the upload is completed; after receiving the abnormal trigger marker, the chest wall patch acquisition terminal is controlled to exit the sleep state and enter the high-fidelity event upload mode; in the high-fidelity event upload mode, the original respiratory sound audio stream is continuously uploaded through the wireless communication module until the abnormal trigger marker disappears and then it returns to the low-power inspection mode.

[0012] Furthermore, the specific steps for determining the fading of the abnormal trigger marker are as follows: In high-fidelity event upload mode, the abnormal gating module continuously receives the respiratory sound feature vector and adhesion quality features within the subsequent monitoring time window; it regenerates the corrected respiratory sound feature vector according to the dual-baseline abnormal gating algorithm for adhesion disturbance correction; based on the degree of deviation of the regenerated corrected respiratory sound feature vector from the individualized respiratory sound baseline template and the respiratory sound baseline parameters of the same age group, it recalculates the abnormal gating value; when the abnormal gating value is lower than the fallback threshold and continues to meet the preset fallback time, it is determined that the abnormal trigger marker has faded; based on the determination result of the fading of the abnormal trigger marker, the hierarchical transmission module controls the chest wall patch acquisition terminal to fall back from the high-fidelity event upload mode to the low-power inspection mode.

[0013] Furthermore, it also includes an event structuring module. The specific steps of the event structuring module for abnormal segment structuring of the original respiratory sound audio stream are as follows: Based on the abnormal occurrence time information corresponding to the abnormal trigger marker, audio segments before and after the abnormal occurrence time information are extracted from the original respiratory sound audio stream to obtain candidate abnormal respiratory sound segments; respiratory sound type identification is performed on the candidate abnormal respiratory sound segments to obtain the abnormal respiratory sound type and abnormal respiratory sound type confidence; the candidate abnormal respiratory sound segments, abnormal respiratory sound type, abnormal respiratory sound type confidence, abnormal gate value, and abnormal occurrence time information are bound to obtain abnormal respiratory sound event data; real abnormal respiratory sound segments are screened based on the abnormal respiratory sound type confidence and abnormal gate value of the abnormal respiratory sound event data; the real abnormal respiratory sound segments are desensitized and annotated, and the desensitized and annotated real abnormal respiratory sound segments are associated with clinical outcome information to obtain case generation data.

[0014] Furthermore, the specific steps for generating structured respiratory sound case data based on case generation data are as follows: Read the anonymized and labeled real abnormal respiratory sound segments, clinical outcome information, and abnormal respiratory sound event data from the case generation data; match symptom records, sign records, and examination / test records within the same time range based on the abnormal occurrence time information in the abnormal respiratory sound event data; temporally correlate the symptom records, sign records, and examination / test records with the anonymized and labeled real abnormal respiratory sound segments to obtain case context data; generate case verification labels based on the case context data and clinical outcome information; combine the anonymized and labeled real abnormal respiratory sound segments, case context data, and case verification labels to form structured respiratory sound case data for pediatric respiratory sound teaching.

[0015] A pediatric chest wall patch-based wireless intelligent respiratory sound monitoring method includes the following steps: detachably connecting the main unit to a disposable patch ring, aligning the pickup window with the hollow area of ​​the disposable patch ring to form a chest wall patch-based acquisition terminal; attaching the chest wall patch-based acquisition terminal to the child's chest wall via the disposable patch ring; acquiring the child's chest wall respiratory sound signal through the hollow area using a respiratory sound acquisition module; processing the child's chest wall respiratory sound signal into a digital respiratory sound signal using a signal preprocessing module; and extracting respiratory sound features from the digital respiratory sound signal using a feature extraction module. Vector and adhesion quality features; a dual-baseline anomaly gating algorithm is used to perform adhesion perturbation correction on the respiratory sound feature vector and adhesion quality features through an anomaly gating module to generate regular monitoring markers or anomaly trigger markers; the hierarchical transmission module controls the chest wall patch acquisition terminal to enter a low-power inspection mode and upload compressed respiratory sound segments according to the regular monitoring markers, and controls the chest wall patch acquisition terminal to enter a high-fidelity event upload mode and upload the original respiratory sound audio stream according to the anomaly trigger markers, and controls the chest wall patch acquisition terminal to return to the low-power inspection mode after the anomaly trigger markers fade.

[0016] The present invention has the following beneficial effects: (1) The pediatric chest wall dressing wireless intelligent respiratory sound monitoring system detachably connects the main unit to the disposable dressing ring, and aligns the hollow area of ​​the disposable dressing ring with the sound pickup window. The respiratory sound acquisition module can collect the respiratory sound signal of the child's chest wall through the hollow area, avoiding the use of vests or chest straps to cover and fix the child's chest wall, reducing the impact of restraint, compression, and pulling on the auscultation points. The disposable dressing ring is replaced separately as a skin contact component, and the main unit is not discarded along with the dressing ring. This is conducive to taking into account hospital infection control, main unit reuse, and consumable cost control, and provides a better foundation for continuous monitoring of pediatric respiratory sounds.

[0017] (2) The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system extracts respiratory sound feature vectors and patch quality features from digital respiratory sound signals simultaneously through the feature extraction module. This allows the abnormal gating module to not only rely on the energy, spectrum, and cepstrum information of the respiratory sound itself when judging changes in respiratory sounds, but also to introduce patch status information such as low-frequency baseline drift, instantaneous energy mutation, spectral quality heart rate variable, and effective energy ratio of the signal. After the patch quality features and respiratory sound feature vectors are bound to the same monitoring time window, they can provide a basis for subsequent patch disturbance correction and reduce false triggers caused by changes in patch tightness, changes in the child's position, or local displacement.

[0018] (3) The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system first generates an individualized respiratory sound baseline template, patch stability benchmark, and respiratory sound baseline parameters of the same age group based on the initial acquisition window through a dual baseline abnormality gating algorithm for patch perturbation correction. Then, it generates a patch perturbation coefficient based on the patch quality characteristics in the subsequent monitoring time window and uses the patch perturbation coefficient to correct the respiratory sound feature vector in the subsequent monitoring time window. The corrected respiratory sound feature vector is then compared with the individualized respiratory sound baseline template and the respiratory sound baseline parameters of the same age group to reduce the feature shift caused by patch perturbation in the subsequent deviation comparison and reduce unnecessary abnormal triggering caused by individual baseline differences and daily patch fluctuations.

[0019] (4) The pediatric chest wall patch wireless intelligent respiratory sound monitoring method drives the hierarchical transmission module to switch working modes through regular monitoring markers and abnormal trigger markers. Under the regular monitoring marker, the chest wall patch acquisition terminal is controlled to enter the low power inspection mode and upload compressed respiratory sound segments. Under the abnormal trigger marker, it switches to the high-fidelity event upload mode and uploads the original respiratory sound audio stream. After the abnormal trigger marker disappears, it returns to the low power inspection mode. In this way, the transmission burden can be reduced during the regular monitoring stage, and the original respiratory sound audio stream can be retained during the abnormal stage, so that the power supply control, abnormal audio retention and continuous monitoring process are linked.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Figure 1 This is a block diagram of a pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to the present invention.

[0022] Figure 2 This is a flowchart of a pediatric chest wall patch-based wireless intelligent respiratory sound monitoring method according to the present invention. Detailed Implementation

[0023] Please see Figure 1 This invention provides a technical solution: a pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system, comprising: a chest wall patch-type acquisition terminal, including a main body and a disposable patch ring; the main body is provided with a pickup window, a respiratory sound acquisition module, a signal preprocessing module, a feature extraction module, and a wireless communication module; the disposable patch ring is detachably connected to the main body and has a hollow area corresponding to the pickup window; the respiratory sound acquisition module acquires the respiratory sound signal of the child's chest wall through the hollow area; the signal preprocessing module processes the respiratory sound signal of the child's chest wall into a digital respiratory sound signal; the feature extraction module extracts respiratory sound feature vectors and patch quality features from the digital respiratory sound signal; and an anomaly gating module is used to perform a dual-baseline anomaly gating algorithm for patch perturbation correction on the respiratory sound feature vectors and patch quality features, including: based on the respiratory sound feature vectors in the initial acquisition window and the patch quality features... The system generates individualized respiratory sound baseline templates, attachment stability benchmarks, and respiratory sound baseline parameters for the same age group based on quality characteristics, the child's age group, and body position information. It generates attachment perturbation coefficients based on attachment quality characteristics within subsequent monitoring time windows, uses these coefficients to correct respiratory sound feature vectors within those windows, and generates routine monitoring markers or abnormal trigger markers based on the deviation of the corrected respiratory sound feature vectors from the individualized respiratory sound baseline templates and the same age group respiratory sound baseline parameters. A tiered transmission module controls the chest wall patch acquisition terminal to enter a low-power inspection mode and upload compressed respiratory sound segments based on routine monitoring markers, and to enter a high-fidelity event upload mode and upload the original respiratory sound audio stream based on abnormal trigger markers. After the abnormal trigger markers fade, the module controls the chest wall patch acquisition terminal to revert to low-power inspection mode.

[0024] In one implementation, the chest wall patch-type data acquisition terminal serves as a sensing device and is connected to an edge computing gateway or a smartphone. The edge computing gateway or smartphone is then connected to a cloud analysis platform. The cloud analysis platform outputs the processing results to the doctor's end, the teaching end, or the parent's end. The doctor's end, the teaching end, or the parent's end displays abnormal events, respiratory sound fragments, or structured respiratory sound case data according to the corresponding usage scenario. In one embodiment, the chest wall dressing-type data acquisition terminal adopts a hospital reusable host-dress separation structure; In an alternative implementation, the chest wall patch-type data acquisition terminal may adopt a fully disposable integrated structure; The signal preprocessing module includes a fourth-order Butterworth bandpass filter unit, an adjustable gain preamplifier unit, and a 16-bit analog-to-digital converter unit. An edge computing gateway or smartphone is equipped with an NLMS adaptive noise reduction module to reduce the impact of environmental noise on the chest wall breath sound signal of the child. The cloud analysis platform can be configured with a multimodal time-series fusion model to perform time alignment processing on breath sound characteristics, physical signs records, and examination and test records.

[0025] In one implementation, the system further includes an event structuring module, which processes the original respiratory sound audio stream according to the abnormal occurrence time information corresponding to the abnormal triggering flag, obtains case generation data, and generates structured respiratory sound case data based on the case generation data.

[0026] Specifically, the attachment and acquisition structure of the chest wall patch-type acquisition terminal is set as follows: The disposable patch is designed with a hollow ring structure, and the hollow area corresponds coaxially with the pickup window on the bottom surface of the main unit. Specifically: In one embodiment, the maximum outer diameter of the disposable dressing ring is set to 40-60mm, the inner diameter is set to 15-25mm, and the overall thickness is controlled to be 0.5-2.0mm, to accommodate the differences in chest size and intercostal space among children of different ages. The hollow area is positioned to correspond with the pickup window on the bottom of the main unit, ensuring a correspondence between the pickup window and the hollow area. This reduces changes in the acquisition position caused by the offset of the pickup window and helps maintain the consistency of breath sound acquisition at the same attachment point.

[0027] The bottom surface of the disposable dressing ring is provided with a medical pressure-sensitive adhesive layer, the middle part of the disposable dressing ring is provided with a medical polyurethane film substrate, and the top surface of the disposable dressing ring is provided with a detachable connection structure that cooperates with the main body of the main unit, specifically as follows: In one embodiment, the medical pressure-sensitive adhesive layer is made of medical acrylate material, and its peel strength is tested according to GB / T 2792 standard to be 1.0-3.0N / 25mm, which takes into account both the adhesion and the ease of removal. The medical polyurethane film substrate has an air permeability of ≥800g / m² / 24h and a thickness of approximately 1.0mm, which helps improve the breathability when children wear it for extended periods. The detachable connection structure on the top surface adopts two optional implementation methods: one is an annular slot interference fit structure, and the other is an N52 grade micro neodymium iron boron magnetic sheet and 430 stainless steel metal ring magnetic attraction fit structure. Both can be used to realize the assembly and disassembly of the main body and the disposable patch.

[0028] The main unit is fixed to the disposable adhesive ring via a detachable connection structure, so that the pickup window corresponds to the hollow area of ​​the disposable adhesive ring, specifically as follows: In one embodiment, the main body adopts an ultra-thin, flat, rounded rectangular structure with an overall size of ≤40mm×30mm×8mm and a weight of ≤20g. The shell is made of medical-grade ABS or covered with medical silicone, with a waterproof rating of ≥IPX4 and no exposed physical interfaces, which facilitates clinical disinfection operations. After fixation, the pickup window and the upper port of the air coupling cavity are aligned, so that the respiratory sound signal of the child's chest wall is transmitted to the internal MEMS microphone through the air coupling cavity.

[0029] After the chest wall patch-type acquisition terminal is attached to the child's chest wall with a disposable patch ring, the hollow area forms an air coupling cavity between the child's chest wall and the pickup window, specifically as follows: In one embodiment, the molding thickness of the air coupling cavity is set to 0.5-2.0 mm, and a medical silicone film layer is provided on the surface of the pickup window. The medical silicone film layer is used to isolate skin secretions and maintain a flexible contact environment on the surface of the pickup window. The air coupling cavity is used to limit the pickup distance between the child's chest wall and the pickup window, reducing the impact of changes in the tightness of the attachment on the consistency of low-frequency respiratory sound acquisition.

[0030] The breath sound acquisition module receives the child's chest wall breath sound signal through the air coupling cavity and transmits the signal to the signal preprocessing module, which specifically involves: In one implementation, the respiratory sound acquisition module uses a Knowles SPH0641LU4H-1 model or equivalent domestic MEMS microphone with a frequency response covering 20Hz-4000Hz, a signal-to-noise ratio ≥62dB, and a sensitivity of -26dBFS±1dB to acquire respiratory sound signals from the chest wall of children of different intensities. The acquired raw analog audio signals are then transmitted to the signal preprocessing module for filtering, amplification, and analog-to-digital conversion.

[0031] In this implementation plan, the feature extraction module generates respiratory sound feature vectors and adhesion quality features simultaneously within the same monitoring time window, enabling the subsequent abnormal gating module to obtain both respiratory sound feature vectors and adhesion quality features at the same time, thereby reducing judgment bias caused by the child's activity, changes in body position, or loosening of the dressing.

[0032] Specifically, the feature extraction module extracts the breath sound feature vector and attaches quality features in the following steps: The digital breath sound signal is segmented according to the monitoring time window to obtain a windowed digital breath sound signal, which is as follows: In one implementation, a 30-minute monitoring time window can be used to equally segment the continuous digital breath sound signal, and the time window can be adaptively adjusted to a 1-minute short window for short-term abnormal monitoring scenarios. The segmentation process retains the timestamps of each window, which facilitates subsequent time alignment of multimodal data and reduces the impact of signal segmentation misalignment on feature extraction.

[0033] The respiratory frequency, energy envelope, spectral centroid, zero-crossing rate, and Mel-frequency cepstral coefficients are extracted from the windowed digital respiratory sound signal to form a respiratory sound feature vector, which is as follows: In one embodiment, the above features are used to characterize the main spectral and temporal variations of pediatric breath sounds, wherein the respiratory rate reflects the child's basic respiratory state, the energy envelope characterizes the overall intensity variation of breath sounds, the spectral centroid distinguishes the distribution of high and low frequency breath sounds, the zero-crossing rate identifies intermittent abnormal breath sounds, the 13-dimensional Mel frequency cepstral coefficient characterizes the spectral details of breath sounds, and the combination of multiple features forms a breath sound feature vector, which is used to characterize the main features of common breath sound types such as normal breath sounds, wet rales, and wheezing.

[0034] The low-frequency baseline drift, instantaneous energy mutation, spectral quality heartbeat variable, and effective energy ratio of the windowed digital breath sound signal are extracted to form the adhesion quality characteristics, which are as follows: In one implementation, this set of features is used to characterize acquisition interference caused by child activity, body position shift, loose dressing, and device displacement, and together with the respiratory sound change features, participates in subsequent abnormal gating; Low-frequency baseline drift reflects baseline offset caused by slight equipment displacement; instantaneous energy mutation reflects interference from sudden changes in body position; spectral quality heartbeat variable reflects spectral fluctuations caused by changes in acquisition status; signal effective energy ratio reflects the tightness of adhesion; and multiple types of adhesion quality characteristics jointly characterize the real-time adhesion status.

[0035] By binding the breath sound feature vector and the adhesion quality feature according to the same monitoring time window, the breath sound window feature data with adhesion status markers is obtained, specifically as follows: In one implementation, a single monitoring time window is used as the unique binding unit, so that each set of breath sound change features corresponds to the attachment status data within the same monitoring time window, reducing the time misalignment between the attachment quality features and the breath sound feature vector, and providing corresponding data for subsequent disturbance correction and anomaly detection.

[0036] The respiratory sound window feature data with attached status markers is sent to the anomaly gating module, specifically as follows: In one implementation, feature data is transmitted within the terminal via a local high-speed bus, eliminating the need for wireless forwarding and reducing the latency and packet loss caused by wireless forwarding. This allows the anomaly gating module to obtain the respiratory sound feature vector and attachment quality features within the same monitoring time window.

[0037] In this implementation scheme, the anomaly gating module first uses the initial acquisition window to form an attachment stability benchmark, and then uses the attachment quality features within the subsequent monitoring time window to generate an attachment disturbance coefficient, so that the breath sound feature vector is corrected for attachment disturbance before entering the dual baseline deviation calculation.

[0038] Specifically, the steps for generating and correcting the breath sound feature vector within the subsequent monitoring time window for attaching the perturbation coefficient are as follows: The attachment quality characteristics within the initial acquisition window are obtained, and the stable distributions of low-frequency baseline drift, instantaneous energy mutation, spectral quality center of gravity variable, and effective signal energy ratio within the initial acquisition window are calculated to generate an attachment stability benchmark, specifically: In one implementation, the first two hours after the child puts on the device are set as the initial data collection window. This period is used to collect the initial respiratory sounds and initial adhesion status of the child after wearing the device, which will serve as a reference for subsequent monitoring. The mean and standard deviation of various adhesion quality characteristics within this window are statistically analyzed to form a standardized adhesion stability benchmark that fits the individual wearing status of the child, avoiding benchmark differences caused by different children's chest circumferences and wearing positions.

[0039] The adhesion quality characteristics within subsequent monitoring time windows are obtained, and the offset of these characteristics relative to the adhesion stability benchmark is calculated. Specifically: In one implementation, the mean deviation method is used to calculate the offset of the real-time attachment features from the stability benchmark, and the offset of the four types of attachment features is quantified respectively to reduce the impact of fluctuations in a single attachment quality feature on the attachment disturbance coefficient and reflect the changes in attachment disturbance.

[0040] The attachment perturbation coefficient is generated based on the offset amplitude. The attachment perturbation coefficient increases with increasing low-frequency baseline drift, instantaneous energy mutation, and spectral centroid variable, and increases with decreasing effective signal energy percentage. Specifically: In one implementation, the attachment perturbation coefficient in the 0-1 range is generated by a multi-feature weighted fusion algorithm. The larger the perturbation amplitude and the lower the effective signal ratio, the larger the attachment perturbation coefficient. The attachment perturbation coefficient is updated according to the changes in the child's activity status, reducing the reliance on a single fixed threshold.

[0041] The energy envelope and spectral centroid of the breath sound feature vector within the subsequent monitoring time window are compensated using the attachment perturbation coefficient to obtain the corrected breath sound feature vector, which is as follows: In one implementation, adaptive compensation is performed only on the energy envelope and spectral centroid of the breath sound feature vector that are susceptible to attachment perturbations, while preserving anti-interference features such as breath frequency and Mel frequency cepstral coefficients. The correction formula is as follows: ; in, These are the corrected characteristic parameters. These are the original feature parameters. To attach the disturbance coefficient, The preset compensation strength coefficient, The value of is greater than 0 and less than 1. Pre-calibrate according to the child's age group, wearing location, or monitoring scenario.

[0042] The process of inputting the corrected breath sound feature vector into the individualized breath sound baseline template and calculating the deviation of the breath sound baseline parameters of the same age group is as follows: In one implementation, the corrected breath sound feature vector reduces device wearing disturbance interference and reduces the impact of attachment disturbance on breath sound variation characteristics, providing a corrected data basis for subsequent dual baseline deviation calculation and anomaly detection.

[0043] In this implementation, the anomaly gating module compares the corrected respiratory sound feature vector with the individualized respiratory sound baseline template and the respiratory sound baseline parameters of the same age group, and judges abnormal changes by continuous accumulation, thereby reducing the impact of occasional fluctuations within a single monitoring time window on the anomaly triggering results.

[0044] Specifically, the steps for generating regular monitoring tags or anomaly trigger tags are as follows: The individual baseline deviation is obtained based on the Mahalanobis distance between the corrected breath sound feature vector and the individualized breath sound baseline template, specifically as follows: In one implementation, the individualized breath sound baseline template is a set of breath sound features within the initial acquisition window of the child. Mahalanobis distance can eliminate the influence of feature dimensions. The difference between the real-time feature vector after quantization correction and the initial breath sound state of the child is calculated using the following formula: ; in, This is the corrected real-time feature vector. The baseline mean of individualized breath sounds. This is the baseline covariance matrix of individualized breath sounds.

[0045] The baseline deviation of the population is obtained by using the Mahalanobis distance between the corrected breath sound feature vector and the baseline parameters of the breath sounds of the same age group. Specifically: In one implementation, the baseline parameters of respiratory sounds for the same age group are pre-stored in five age groups: 0-1 years, 1-3 years, 3-6 years, 6-12 years, and 12-18 years. Simultaneously, three body positions are distinguished: supine, lateral, and sitting, to accommodate differences in respiratory sounds across different age groups and positions. The formula for calculating the baseline deviation of the group is as follows: ; in, This represents the baseline mean for the same age group. This represents the baseline covariance matrix for the same age group.

[0046] Individual baseline deviations and group baseline deviations are input into a dual baseline deviation fusion function to obtain anomaly gating values, specifically: In one implementation, a weighted fusion method is used to balance individual specificity and group universality. The fusion formula is as follows: ; in, The weighting coefficient, ranging from 0.6 to 0.8, gives higher weight to individual baseline deviations in the fusion results, reducing the problem of insufficient adaptation to individual differences when only using population baseline parameters.

[0047] The abnormal gate values ​​within the continuous monitoring time window are accumulated by sliding according to time sequence to obtain the abnormal gate duration, which is as follows: In one implementation, five consecutive time windows are set as sliding accumulation units to perform time-series accumulation calculations on abnormal threshold values, thereby reducing the impact of instantaneous noise or accidental fluctuations within a single monitoring time window on the abnormal triggering results.

[0048] An anomaly trigger flag is generated when the anomaly gating duration reaches the anomaly trigger condition; a normal monitoring flag is generated when the anomaly gating duration does not reach the anomaly trigger condition. Specifically: In one implementation, a tiered trigger threshold is set, with a yellow warning threshold and a red emergency threshold set in layers. When the duration of the abnormal gating exceeds the basic threshold, an abnormal trigger mark is generated; if it is below the threshold, a regular monitoring mark is generated, and an abnormal trigger result of the corresponding level is generated.

[0049] In this implementation plan, the hierarchical transmission module uses the regular monitoring flag or abnormal trigger flag generated by the abnormal gating module as the basis for switching transmission modes, so that the chest wall patch acquisition terminal reduces the communication burden in the normal state and retains the original respiratory sound audio stream in the abnormal state.

[0050] Specifically, the transmission control steps of the hierarchical transmission module are as follows: After receiving the routine monitoring markers, the chest wall patch-type data acquisition terminal is controlled to collect digital breath sound signals according to the inspection and collection cycle, specifically as follows: In one implementation, the low-power inspection mode sets an inspection and acquisition cycle of 60-120 minutes, and collects 30-60 seconds of respiratory sound signals at a time, taking into account both routine inspection and acquisition and low-power operation of the equipment, which is suitable for continuous monitoring scenarios in pediatric inpatient settings.

[0051] The digital breath sound signals collected during the inspection cycle are compressed and encoded to obtain compressed breath sound segments, specifically: In one implementation, the OPUS 16kbps compression encoding format is used to compress the regular breath sound segments, reducing the amount of data transmission while preserving the main breath sound characteristics, and adapting to the low-power asynchronous transmission mode.

[0052] The compressed breath sound clip is uploaded via a wireless communication module, and after the upload is complete, the chest wall patch-type acquisition terminal is controlled to enter a sleep state. Specifically: In one implementation, the wireless communication module adopts the Star Flash SLE asynchronous burst transmission mode to complete the batch data upload. After the upload is completed, the connection is disconnected and the module enters a deep sleep state, which helps to reduce the communication power consumption during the regular monitoring phase and extend the working time of the chest wall patch-type data acquisition terminal.

[0053] Upon receiving an abnormal trigger flag, the chest wall patch-type data acquisition terminal exits sleep mode and enters high-fidelity event upload mode, specifically as follows: In one implementation, after an anomaly is triggered, the control chest wall patch-type acquisition terminal is woken up, the communication module switches from sleep asynchronous mode to continuous connection mode, stops compressing and encoding the original respiratory sound audio stream during the abnormal event, and retains the original respiratory sound information during the abnormal event.

[0054] In high-fidelity event upload mode, the original breathing sound audio stream is continuously uploaded via wireless communication module until the abnormal trigger flag fades, after which it switches back to low-power inspection mode, specifically as follows: In one implementation, the original respiratory sound audio stream is uploaded in a 16kHz / 16bit sampling format in the high-fidelity event upload mode. The upload is continuously collected for 1-5 minutes, and the original respiratory sound information during the abnormal event is retained. After the abnormal trigger mark fades, the mode switches back to low-power inspection mode, taking into account both power consumption control during the regular monitoring phase and audio retention during the abnormal event phase.

[0055] In this implementation scheme, the determination of abnormal trigger mark fading adopts the dual-baseline abnormality gating algorithm of attachment perturbation correction, so that the entry and exit of high-fidelity event upload mode adopt the same determination logic, reducing frequent mode switching.

[0056] Specifically, the steps for determining the expiration of the exception trigger flag are as follows: In high-fidelity event upload mode, the anomaly gating module continuously receives respiratory sound feature vectors and adhesion quality features within subsequent monitoring time windows, specifically: In one implementation, the feature extraction time window is shortened in the high-fidelity event upload mode, and respiratory sound feature vectors and attachment quality features are continuously collected to track abnormal state changes.

[0057] The corrected breath sound feature vector is regenerated according to the dual-baseline anomaly gating algorithm for attachment perturbation correction, specifically as follows: In one implementation, the dual-baseline anomaly gating algorithm for attachment perturbation correction is reused to ensure that the anomaly triggering and anomaly triggering flag fading adopt the same judgment rule, thereby avoiding judgment errors caused by the switching of judgment rules.

[0058] Based on the degree of deviation of the regenerated corrected breath sound feature vector from the individualized breath sound baseline template and the breath sound baseline parameters of the same age group, the abnormality gate value is recalculated, specifically as follows: In one implementation, an anomaly gate value is updated to track changes in anomaly states.

[0059] When the anomaly threshold value is lower than the fallback threshold and continues to meet the preset fallback time, the anomaly trigger flag is determined to have faded. Specifically: In one implementation, the preset drop threshold is 0.5 times the base trigger threshold, and the preset drop duration is 3-5 consecutive time windows to reduce frequent mode switching caused by abnormal short-term fluctuations.

[0060] Based on the fading of abnormal trigger markers, the hierarchical transmission module controls the chest wall patch-type data acquisition terminal to switch back from high-fidelity event upload mode to low-power inspection mode. Specifically: In one implementation, the mode fallback process continues the routine inspection and data collection task without rebinding the device, reducing the impact of mode switching on the continuous monitoring process.

[0061] In this implementation plan, the original respiratory sound audio stream is intercepted, identified, bound, and filtered after an anomaly is triggered, so that the abnormal event can not only serve as a real-time early warning result, but also form structured data for subsequent archiving, review, and teaching use.

[0062] Specifically, it also includes an event structuring module. The specific steps of the event structuring module for abnormal segments of the original breath sound audio stream are as follows: Based on the abnormal occurrence time information corresponding to the abnormal trigger marker, audio segments before and after the abnormal occurrence time information are extracted from the original breath sound audio stream to obtain candidate abnormal breath sound segments, specifically: In one implementation, original audio segments of 60 seconds before and after the occurrence of the abnormal event are extracted to cover the changes in breathing sounds before and after the occurrence of the abnormal event, thereby reducing the possibility of incomplete extraction of abnormal segments.

[0063] The candidate abnormal breath sound segments are subjected to breath sound type identification to obtain the abnormal breath sound type and the confidence level of the abnormal breath sound type, which are as follows: In one implementation, a pre-trained CNN model or Transformer model is used to identify abnormal respiratory sound types such as coarse wet rales, medium wet rales, fine wet rales, high and low pitch wheezing, snoring, and decreased or absent breath sounds, and outputs the corresponding confidence scores.

[0064] By binding candidate abnormal breath sound segments, abnormal breath sound types, abnormal breath sound type confidence levels, abnormal gating values, and abnormal occurrence time information, abnormal breath sound event data is obtained, which is as follows: In one implementation, the audio segment, recognition result, algorithm parameters, and event time are structured and bound together using the timestamp as a unique index to form abnormal breathing sound event data, which facilitates subsequent screening, archiving, and case generation.

[0065] The actual abnormal breath sound segments are filtered based on the confidence level of the abnormal breath sound type and the abnormality gate value of the abnormal breath sound event data. Specifically: In one implementation, a dual screening condition is set: segments with an abnormal breath sound type confidence level that reaches a preset confidence threshold and an abnormality gate value that reaches a preset gate threshold are identified as real abnormal breath sound segments. Invalid abnormal data caused by model misidentification and slight perturbation are eliminated, thereby improving the reliability of screening real abnormal breath sound segments.

[0066] Real abnormal breath sound fragments were desensitized and annotated, and then the desensitized real abnormal breath sound fragments were correlated with clinical outcome information to obtain case generation data, which is as follows: In one implementation, the patient identity information corresponding to the real abnormal breath sound fragments is desensitized by removing identity information such as name, hospital number, and ID number, while retaining clinical information related to case generation. At the same time, the start and end time, abnormality type, and abnormality degree of the abnormal fragments are marked and associated with the corresponding clinical outcome information.

[0067] The specific steps for generating structured breath sound case data based on case generation data are as follows: The process involves retrieving anonymized and labeled segments of actual abnormal breath sounds, clinical outcome information, and abnormal breath sound event data from the generated case data. Specifically: In one implementation, the filtered abnormal breath sound event data are retrieved in batches and uniformly incorporated into the case generation data source, so that the structured breath sound case data originates from the filtered abnormal breath sound event data.

[0068] Based on the abnormal occurrence time information in the abnormal breath sound event data, symptom records, physical sign records, and examination and test records within the same time range are matched, specifically as follows: In one implementation, the system connects to the hospital's HIS / LIS system via the HL7 FHIR standard interface to match body temperature, heart rate, blood oxygen, PCT, CRP, white blood cell count, and pathogen detection data for the corresponding time period, thereby achieving time alignment of sound-symptom-test data.

[0069] By temporally correlating symptom records, physical sign records, and examination and test records with desensitized and annotated segments of actual abnormal breath sounds, case context data is obtained, specifically: In one implementation, linear interpolation is used to supplement clinical data at different sampling frequencies, unify the time grid, align the breath sound time sequence data with clinical test data, and form case context data.

[0070] Case validation labels are generated based on case context data and clinical outcome information, specifically as follows: In one implementation, a verification label is generated by combining the child's final diagnosis, the improvement / deterioration / stabilization status of the condition, and the discharge record. This label serves as the case verification label in the structured breath sound case data and provides a corresponding clinical outcome reference for the structured breath sound case data.

[0071] The desensitized and annotated real abnormal breath sound fragments, case context data, and case verification labels are combined to form structured breath sound case data for pediatric breath sound teaching, specifically as follows: In one implementation, the final generated structured case includes audio files, waveform spectrum visualization, AI annotation information, clinical diagnosis and treatment summary, and disease outcome verification data, which can be used for pediatric respiratory sound teaching, skills training, and case review.

[0072] In this implementation plan, the pediatric chest wall patch-type wireless intelligent respiratory sound monitoring method corresponds to the chest wall patch-type acquisition terminal, abnormal gating module, and hierarchical transmission module in the above system, so that acquisition, abnormal gating, and hierarchical transmission are performed according to the same data link.

[0073] Please see Figure 2This invention provides a technical solution: a pediatric chest wall patch-type wireless intelligent respiratory sound monitoring method, comprising the following steps: detachably connecting the main unit to a disposable patch ring, aligning the pickup window with the hollow area of ​​the disposable patch ring to form a chest wall patch-type acquisition terminal, and attaching the chest wall patch-type acquisition terminal to the child's chest wall via the disposable patch ring; acquiring the child's chest wall respiratory sound signal through the hollow area using a respiratory sound acquisition module, and processing the signal into a digital respiratory sound signal using a signal preprocessing module; extracting respiratory sound feature vectors and patch quality features from the digital respiratory sound signal using a feature extraction module; and performing anomaly gating on the respiratory sound feature vectors and patch quality features using an anomaly gating module. A dual-baseline anomaly gating algorithm with attached perturbation correction generates regular monitoring markers or anomaly trigger markers. A hierarchical transmission module controls the chest wall patch-type acquisition terminal to enter a low-power inspection mode and upload compressed breath sound segments based on the regular monitoring markers. Based on the anomaly trigger markers, the chest wall patch-type acquisition terminal enters a high-fidelity event upload mode and uploads the original breath sound audio stream. After the anomaly trigger markers fade, the chest wall patch-type acquisition terminal returns to the low-power inspection mode. After generating the original breath sound audio stream, the system can also perform abnormal segment structuring processing on the original breath sound audio stream based on the anomaly occurrence time information corresponding to the anomaly trigger markers to obtain case generation data. Structured breath sound case data is then generated based on the case generation data.

[0074] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0075] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system, characterized in that, include: The chest wall dressing-type data acquisition terminal includes a main unit and a disposable dressing ring. The main unit is equipped with a sound pickup window, a breath sound acquisition module, a signal preprocessing module, a feature extraction module, and a wireless communication module. The disposable dressing ring is detachably connected to the main unit and has a hollow area corresponding to the sound pickup window. The breath sound acquisition module acquires the child's chest wall breath sound signal through the hollow area. The signal preprocessing module processes the child's chest wall breath sound signal into a digital breath sound signal. The feature extraction module extracts the breath sound feature vector and the dressing quality feature from the digital breath sound signal. The anomaly gating module is a dual-baseline anomaly gating algorithm used to perform attachment perturbation correction on the breath sound feature vector and attachment quality features. It includes: generating an individualized breath sound baseline template, attachment stability benchmark, and breath sound baseline parameters of the same age group based on the breath sound feature vector, attachment quality features, age group, and body position information of the child in the initial acquisition window; generating an attachment perturbation coefficient based on the attachment quality features in the subsequent monitoring time window; correcting the breath sound feature vector in the subsequent monitoring time window using the attachment perturbation coefficient; and generating a regular monitoring mark or anomaly trigger mark based on the degree of deviation of the corrected breath sound feature vector from the individualized breath sound baseline template and the breath sound baseline parameters of the same age group. The hierarchical transmission module is used to control the chest wall patch acquisition terminal to enter the low-power inspection mode and upload compressed respiratory sound segments according to the regular monitoring markers, to control the chest wall patch acquisition terminal to enter the high-fidelity event upload mode and upload the original respiratory sound audio stream according to the abnormal trigger markers, and to control the chest wall patch acquisition terminal to return to the low-power inspection mode after the abnormal trigger markers fade.

2. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 1, characterized in that, The attachment and acquisition structure of the chest wall patch-type data acquisition terminal is set as follows: The disposable dressing ring is designed with a hollow ring structure, and the hollow area corresponds coaxially with the pickup window on the bottom of the main unit. The bottom surface of the disposable dressing ring is provided with a medical pressure-sensitive adhesive layer, the middle part of the disposable dressing ring is provided with a medical polyurethane film substrate, and the top surface of the disposable dressing ring is provided with a detachable connection structure that cooperates with the main body. The main unit is fixed to the disposable dressing ring through a detachable connection structure, so that the pickup window corresponds to the hollow area of ​​the disposable dressing ring; After the chest wall patch-type acquisition terminal is attached to the chest wall of the child with a disposable patch, the hollow area forms an air coupling cavity between the child's chest wall and the pickup window. The respiratory sound acquisition module receives the respiratory sound signal of the child's chest wall through the air coupling cavity and transmits the signal to the signal preprocessing module.

3. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 1, characterized in that, The specific steps of the feature extraction module in extracting the respiratory sound feature vector and attaching quality features are as follows: The digital breath sound signal is segmented according to the monitoring time window to obtain a windowed digital breath sound signal; The respiratory frequency, energy envelope, spectral centroid, zero-crossing rate, and Mel frequency cepstral coefficients are extracted from the windowed digital respiratory sound signal to form a respiratory sound feature vector. Low-frequency baseline drift, instantaneous energy mutation, spectral quality heartbeat variable, and effective energy ratio of the signal are extracted from the windowed digital breath sound signal to form adhesion quality characteristics; The respiratory sound feature vector and the adhesion quality feature are bound together according to the same monitoring time window to obtain respiratory sound window feature data with adhesion status markers; Send the respiratory sound window feature data with attached status markers to the anomaly gating module.

4. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 1, characterized in that, The specific steps for generating and correcting the breath sound feature vector within the subsequent monitoring time window using the attached perturbation coefficient are as follows: The attachment quality characteristics within the initial acquisition window are obtained, and the stable distribution of low-frequency baseline drift, instantaneous energy mutation, spectral quality center jump variable, and effective signal energy ratio within the initial acquisition window is calculated to generate an attachment stability benchmark. Obtain the adhesion quality characteristics within the subsequent monitoring time window, and calculate the offset of the adhesion quality characteristics relative to the adhesion stability benchmark within the subsequent monitoring time window. The attachment perturbation coefficient is generated based on the offset amplitude. The attachment perturbation coefficient increases with the increase of low-frequency baseline drift, instantaneous energy mutation and spectral centroid variable, and increases with the decrease of the effective energy ratio of the signal. The energy envelope and spectral centroid of the respiratory sound feature vector within the subsequent monitoring time window are compensated by the attachment perturbation coefficient to obtain the corrected respiratory sound feature vector; The process involves inputting the corrected respiratory sound feature vector into the individualized respiratory sound baseline template and calculating the deviation of the respiratory sound baseline parameters of the same age group.

5. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 1, characterized in that, The specific steps for generating regular monitoring tags or anomaly trigger tags are as follows: The individual baseline deviation is obtained by using the Mahalanobis distance between the corrected respiratory sound feature vector and the individualized respiratory sound baseline template. The baseline deviation of the population is obtained by the Mahalanobis distance between the corrected respiratory sound feature vector and the baseline parameters of respiratory sounds of the same age group. Individual baseline deviations and group baseline deviations are input into a dual baseline deviation fusion function to obtain anomaly gating values; The abnormal gate value within the continuous monitoring time window is accumulated by sliding according to the time sequence to obtain the abnormal gate duration. An abnormal trigger flag is generated when the abnormal gating duration reaches the abnormal trigger condition, and a normal monitoring flag is generated when the abnormal gating duration does not reach the abnormal trigger condition.

6. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 1, characterized in that, The specific steps for transmission control of the hierarchical transmission module are as follows: After receiving the routine monitoring markers, control the chest wall patch-type acquisition terminal to collect digital respiratory sound signals according to the inspection and acquisition cycle; The digital respiratory sound signals collected during the inspection cycle are compressed and encoded to obtain compressed respiratory sound segments. The compressed breath sound segment is uploaded via a wireless communication module, and the chest wall patch acquisition terminal is put into sleep mode after the upload is completed. Upon receiving an abnormal trigger flag, the chest wall patch-type data acquisition terminal is controlled to exit the sleep state and enter the high-fidelity event upload mode. In high-fidelity event upload mode, the original breathing sound audio stream is continuously uploaded via wireless communication module until the abnormal trigger mark fades and then it returns to low-power inspection mode.

7. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 1, characterized in that, The specific steps for determining the expiration of the abnormal trigger marker are as follows: In high-fidelity event upload mode, the anomaly gating module continuously receives respiratory sound feature vectors and adhesion quality features within the subsequent monitoring time window; The corrected breath sound feature vector is regenerated according to the dual baseline anomaly gating algorithm with attachment perturbation correction; Based on the degree of deviation of the regenerated corrected respiratory sound feature vector from the individualized respiratory sound baseline template and the respiratory sound baseline parameters of the same age group, the abnormal gating value is recalculated. When the abnormal threshold value is lower than the fallback threshold and the preset fallback time is met, the abnormal trigger flag is determined to fade. Based on the determination result of the abnormal trigger marker fading, the hierarchical transmission module controls the chest wall patch-type acquisition terminal to switch back from high-fidelity event upload mode to low-power inspection mode.

8. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 1, characterized in that, It also includes an event structuring module, which performs the following specific steps on abnormal segments of the original breath sound audio stream for structuring: Based on the abnormal occurrence time information corresponding to the abnormal trigger flag, audio segments before and after the abnormal occurrence time information are extracted from the original respiratory sound audio stream to obtain candidate abnormal respiratory sound segments. The candidate abnormal breath sound segments are identified to obtain the abnormal breath sound type and the confidence level of the abnormal breath sound type. By binding candidate abnormal breath sound segments, abnormal breath sound types, abnormal breath sound type confidence levels, abnormal gating values, and abnormal occurrence time information, abnormal breath sound event data can be obtained. Filter the actual abnormal breath sound segments based on the confidence level of the abnormal breath sound type and the abnormal gating value of the abnormal breath sound event data; Desensitized and annotated real abnormal breath sound segments, and then linked the desensitized and annotated real abnormal breath sound segments with clinical outcome information to obtain case generation data.

9. The pediatric chest wall patch-type wireless intelligent respiratory sound monitoring system according to claim 8, characterized in that, The specific steps for generating structured breath sound case data based on case generation data are as follows: Read the desensitized and annotated real abnormal breath sound fragments, clinical outcome information, and abnormal breath sound event data from the case generation data; Based on the abnormal occurrence time information in the abnormal breath sound event data, match symptom records, physical sign records, and examination and test records within the same time range; By temporally correlating symptom records, physical sign records, and examination and test records with desensitized and labeled real abnormal breath sound segments, case context data is obtained. Generate case validation labels based on case context data and clinical outcome information; By combining anonymized and labeled real abnormal breath sound fragments, case context data, and case verification labels, structured breath sound case data for pediatric breath sound teaching is formed.

10. A pediatric chest wall patch-based wireless intelligent respiratory sound monitoring method, using the pediatric chest wall patch-based wireless intelligent respiratory sound monitoring system according to any one of claims 1-9, characterized in that, Includes the following steps: The main unit is detachably connected to the disposable dressing ring, so that the sound pickup window corresponds to the hollow area of ​​the disposable dressing ring, forming a chest wall dressing acquisition terminal, and the chest wall dressing acquisition terminal is attached to the chest wall of the child through the disposable dressing ring. Breath sound signals from the child's chest wall were acquired through the hollow region using a breath sound acquisition module, and then processed into digital breath sound signals using a signal preprocessing module. The feature extraction module extracts respiratory sound feature vectors and attachment quality features from digital respiratory sound signals. A dual-baseline anomaly gating algorithm, which performs attachment perturbation correction on the breath sound feature vector and attachment quality feature through an anomaly gating module, generates regular monitoring markers or anomaly trigger markers. The hierarchical transmission module controls the chest wall patch acquisition terminal to enter a low-power inspection mode and upload compressed respiratory sound segments based on the regular monitoring markers. Based on the abnormal trigger markers, it controls the chest wall patch acquisition terminal to enter a high-fidelity event upload mode and upload the original respiratory sound audio stream. After the abnormal trigger markers fade, it controls the chest wall patch acquisition terminal to return to the low-power inspection mode.

Images