Data analysis and early warning method and system for intelligent sputum aspirator

Abstract

The invention discloses a data analysis and early warning method and system for an intelligent sputum aspirator, and relates to the technical field of sputum aspirates.The data analysis and early warning method for the intelligent sputum aspirator comprises the following steps that S1, an acquisition request sent by a user side is received, an acquisition execution device is determined from a corresponding acquisition subsystem according to the acquisition request, and the acquisition execution device is connected with the user side; establishing a communication channel with the acquisition execution equipment; s2, receiving original collection data sent by the collection execution equipment through a communication channel, processing the original collection data to obtain denoised data and normalized data, and taking the original collection data, the denoised data and the normalized data as multi-source data; and S3, performing multi-level analysis on the multi-source data to obtain a multi-level early warning result, and sending the multi-level early warning result to a user side. According to the invention, the problems of incomplete data acquisition, non-fine processing and inaccurate early warning in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of suction device technology, and in particular to a data analysis and early warning method and system for intelligent suction devices. Background Technology

[0002] In the application of intelligent suction devices, effective monitoring and accurate early warning of patient suction-related data are crucial for ensuring patient safety and improving the quality of medical care. However, existing technologies have several shortcomings in data processing and early warning for intelligent suction devices, as follows:

[0003] (1) Existing early warning systems often struggle to collect multi-dimensional data during sputum suction in a comprehensive and efficient manner. Patient pressure data, sputum images, sputum flow, respiratory sound data, suction frequency, body temperature, and respiratory rate are often scattered across different acquisition subsystems or devices. The lack of a unified and flexible acquisition scheduling mechanism makes it difficult to quickly establish a stable communication channel to acquire data. As a result, data collection is often untimely and incomplete, failing to provide a reliable foundation for subsequent accurate analysis.

[0004] (2) Traditional early warning methods do not systematically preprocess the collected data, resulting in poor raw data quality that can easily affect the accuracy of the analysis results. Furthermore, in the early warning stage, traditional methods often rely on simple threshold judgments for single data points, failing to perform multi-level hierarchical analysis of multi-source data. They also cannot simultaneously provide real-time and predictive early warning functions, making it difficult to accurately identify abnormal equipment operation (such as abnormal equipment pressure or flow) and potential health risks to patients. Consequently, they fail to provide comprehensive and valuable auxiliary decision-making support for medical staff, increasing medical and nursing risks and reducing nursing efficiency. Summary of the Invention

[0005] The purpose of this application is to provide a data analysis and early warning method and system for intelligent suction devices, which solves the problems of incomplete data collection, imprecise processing, and inaccurate early warning in the existing technology. It can provide medical staff with more reliable equipment status monitoring and patient health risk prediction, assist medical staff in taking timely intervention measures, improve the safety and effectiveness of suction care, reduce medical risks, and optimize the allocation of medical resources.

[0006] To achieve the above objectives, this application provides a data analysis and early warning method for an intelligent suction device, comprising the following steps: S1: Receiving a collection request sent by a user terminal, determining the collection execution device from the corresponding collection subsystem according to the collection request, and establishing a communication channel with the collection execution device; wherein, the collection request includes at least: patient medical data, initial vital sign data, and user terminal ID; S2: Receiving raw collection data sent by the collection execution device through the communication channel, processing the raw collection data to obtain denoised data and normalized data, and using the raw collection data, denoised data, and normalized data as multi-source data; wherein, the raw collection data includes at least: pressure data, sputum image, sputum flow, respiratory sound data, suctioning frequency, patient body temperature, and patient respiratory rate; S3: Performing multi-level analysis on the multi-source data to obtain multi-level early warning results, and sending the multi-level early warning results to the user terminal, wherein the multi-level early warning results include at least: real-time early warning results and predictive early warning results.

[0007] As described above, the sub-steps for determining the acquisition execution device from the corresponding acquisition subsystem according to the acquisition request and establishing a communication channel with the acquisition execution device are as follows: S11: Traverse and analyze the pre-built device database according to the acquisition request to obtain device recommendation information, and send the device recommendation information to the user terminal; wherein, the device recommendation information includes at least: recommended device code and recommended device parameters; S12: Receive the device usage information sent by the user terminal after referring to the device recommendation information, wherein, the device usage information includes at least: device code, multiple parameter configuration items, and the parameter value corresponding to each parameter configuration item; S13: Determine that the device corresponding to the device code is the acquisition execution device, and establish a communication channel with the acquisition execution device.

[0008] As described above, the sub-steps for processing the original collected data to obtain denoised and normalized data are as follows: S21: Use wavelet transform algorithm to denoise the pressure data to obtain denoised pressure data; normalize the denoised pressure data to obtain normalized pressure data; S22: Use Kalman filter method to denoise the sputum flow rate to obtain denoised sputum flow rate; normalize the denoised sputum flow rate to obtain normalized sputum flow rate; S23: Use filtering algorithm to denoise the respiratory sound data to obtain denoised respiratory sound data.

[0009] As described above, the sub-steps for performing multi-level analysis on multi-source data to obtain multi-level early warning results are as follows: S31: Perform real-time analysis on multi-source data to obtain real-time early warning results, wherein the real-time early warning results include at least: real-time early warning results for equipment and real-time early warning results for health; S32: Perform predictive analysis on multi-source data to obtain predictive early warning results, wherein the predictive early warning results include at least: pressure prediction risk results, flow prediction risk results, and health prediction risk results.

[0010] As described above, the sub-steps for real-time analysis of multi-source data to obtain real-time early warning results are as follows: S311: Perform equipment early warning analysis based on the denoised data from the multi-source data to generate real-time equipment early warning results, wherein the real-time equipment early warning results are either risky or risk-free; S312: Perform health early warning analysis based on the denoised data and the original collected data from the multi-source data to generate real-time health early warning results, wherein the real-time health early warning results are either risky or risk-free.

[0011] As described above, the sub-steps for generating real-time equipment warning results based on denoised data from multi-source data are as follows: S3111: Calculate the pressure fluctuation entropy value based on the denoised pressure data in the denoised data, judge the pressure fluctuation entropy value using the pressure fluctuation threshold, and generate a real-time equipment warning result. If the pressure fluctuation entropy value is greater than the pressure fluctuation threshold, it is determined that there is a risk of pipeline blockage, and the generated real-time equipment warning result is "risky"; if the pressure fluctuation entropy value is less than or equal to the pressure fluctuation threshold, proceed to S3113; S3112: Calculate the flow fluctuation entropy value based on the denoised sputum flow rate in the denoised data, judge the flow fluctuation entropy value using the flow fluctuation threshold, and generate a real-time equipment warning result. If the flow fluctuation entropy value is greater than the flow fluctuation threshold, it is determined that the equipment has poor operational stability. The generated real-time device warning result is "risky"; if the flow fluctuation entropy value is less than or equal to the flow fluctuation threshold, then execute S3113; S3113: Set the analysis period. During the analysis period, extract the local maximum value from the denoised pressure data as the pressure peak value, and extract the local minimum value from the denoised sputum flow as the flow valley value; use the difference between the pure value of the pressure peak value and the pure value of the flow valley value as the matching degree of the peak and valley values, and use the preset matching degree threshold to judge the matching degree of the peak and valley values, and generate the real-time device warning result. If the matching degree of the peak and valley values ​​is greater than the preset matching degree threshold, it is determined that the suction pump power is abnormal, and the generated real-time device warning result is "risky". If the matching degree of the peak and valley values ​​is less than or equal to the preset matching degree threshold, the generated real-time device warning result is "no risk".

[0012] As described above, the sub-steps for generating real-time health warning results by performing health warning analysis based on the denoised data and original collected data from multi-source data are as follows: S3121: Calculate the respiratory health risk value based on a multi-factor linear regression model; S3122: Analyze the respiratory health risk value using a preset respiratory health risk threshold to generate a real-time health warning result. If the respiratory health risk value is greater than the preset respiratory health risk threshold, the generated real-time health warning result is considered risky; if the respiratory health risk value is less than or equal to the preset respiratory health risk threshold, the generated real-time health warning result is considered risk-free.

[0013] As described above, the sub-steps for predictive analysis of multi-source data to obtain predictive warning results are as follows: S321: Input the normalized pressure data from the normalized data into a pre-trained pressure prediction model. The pressure prediction model outputs a pressure prediction result. The pressure prediction result is judged using a preset pressure risk threshold to generate a pressure prediction risk result. If the pressure prediction result is greater than the preset pressure risk threshold, the generated pressure prediction risk result is risky; if the pressure prediction result is less than or equal to the preset pressure risk threshold, the generated pressure prediction risk result is risk-free. S322: Input the normalized sputum flow rate from the normalized data into a pre-trained flow prediction model. The flow prediction model outputs a flow prediction result. The flow prediction result is judged using a preset flow risk threshold to determine the risk of the sputum flow rate. The prediction results are judged to generate a flow prediction risk result. If the flow prediction result is greater than the preset flow risk threshold, the generated flow prediction risk result is risky; if the flow prediction result is less than or equal to the preset flow risk threshold, the generated flow prediction risk result is risk-free. S323: The sputum image features, historical vital sign data, pressure prediction results, and flow prediction results are input into the pre-built health prediction model. The health prediction model outputs a health prediction result. The health prediction result is judged using the preset health risk threshold to generate a health prediction risk result. If the health prediction result is greater than the preset health risk threshold, the generated health prediction risk result is risky; if the health prediction result is less than or equal to the preset health risk threshold, the generated health prediction risk result is risk-free.

[0014] As shown above, the pre-trained stress prediction model is a stress prediction model based on a bidirectional long short-term memory network; the pre-trained traffic prediction model is a traffic prediction model based on a bidirectional long short-term memory network; and the pre-built health prediction model is a health prediction model based on a support vector machine and a generative adversarial network.

[0015] This application also provides a data analysis and early warning system for intelligent suction devices, comprising: an analysis and early warning center and at least one user terminal, each user terminal corresponding to at least one data acquisition subsystem; wherein, the analysis and early warning center is used to execute the above-described data analysis and early warning method for intelligent suction devices; the user terminal is used to send data acquisition requests and device information; and to receive multi-level early warning results; the data acquisition subsystem supports the establishment of a communication channel between the analysis and early warning center and a selected data acquisition execution device, and the data acquisition execution device sends raw data to the analysis and early warning center through the communication channel.

[0016] The data analysis and early warning method and system for intelligent suction devices proposed in this application solves the problems of incomplete data collection, imprecise processing, and inaccurate early warning in existing technologies. It can provide medical staff with more reliable equipment status monitoring and patient health risk prediction, assisting medical staff to take timely intervention measures, improving the safety and effectiveness of suction care, reducing medical risks, and optimizing the allocation of medical resources. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 A schematic diagram of a data analysis and early warning system for an intelligent suction device;

[0019] Figure 2 A schematic diagram of one embodiment of a closed suction device system;

[0020] Figure 3 This is a flowchart of one embodiment of a data analysis and early warning method for a smart suction device. Detailed Implementation

[0021] 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, not all, of the embodiments of the present invention. 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.

[0022] like Figure 1As shown, this application provides a data analysis and early warning system for an intelligent suction device, comprising: an analysis and early warning center 1 and at least one user terminal 2, each user terminal 2 corresponding to at least one data acquisition subsystem 3. The analysis and early warning center 1 is used to execute the data analysis and early warning method for the intelligent suction device described below. User terminal 2: used to send data acquisition requests and device information; used to receive multi-level early warning results. Data acquisition subsystem 3: supports the establishment of a communication channel between the analysis and early warning center 1 and a selected data acquisition execution device, through which the data acquisition execution device sends raw data to the analysis and early warning center 1.

[0023] Furthermore, the analysis and early warning center 1 includes at least: a communication unit, a data processing unit, a multi-level early warning analysis unit, and a storage unit.

[0024] The communication unit is used to receive collection requests sent by the user terminal 2, determine the collection execution device from the corresponding collection subsystem 3 according to the collection request, and establish a communication channel with the collection execution device; receive the raw collection data sent by the collection execution device through the communication channel; and send the multi-level early warning results to the user terminal 2.

[0025] Data processing unit: Used to process the raw collected data to obtain denoised data and normalized data, and to treat the raw collected data, denoised data and normalized data as multi-source data.

[0026] Multi-level early warning analysis unit: Performs multi-level analysis on multi-source data to obtain multi-level early warning results.

[0027] Storage unit: Used to store the device database; used to store preset respiratory function indicators, as well as normal reference ranges for circulatory and basal state indicators.

[0028] The device database comprises multiple sub-device databases, each corresponding to a user ID table. Each sub-device database includes: various device data packages, each device data package corresponding to a standard population attribute; each device data package includes: various sub-device data packages, each sub-device data package corresponding to a device code, a standard symptom characteristic data package, and a standard symptom severity table; each sub-device data package includes multiple standard device parameters, each standard device parameter corresponding to a standard severity range; and each standard device parameter includes: multiple parameter settings, each parameter setting corresponding to a recommended parameter range.

[0029] The parameter settings should include at least: negative pressure parameters, suctioning frequency, suctioning time and interval, humidification parameters, and sputum image acquisition frequency.

[0030] Furthermore, the pre-constructed standard disease severity table includes at least: multiple disease severity levels, one disease severity level corresponding to one disease severity value, and a standard disease severity feature data package.

[0031] The specific value for the severity of the illness is set according to the actual situation; the more severe the illness, the higher the severity value.

[0032] The specific values ​​for the recommended parameter range are set based on practical experience and research data.

[0033] Specifically, the types and quantities of equipment in the data acquisition subsystems vary across different institutions or hospitals. Therefore, a device database is used to store equipment information for each institution or hospital. The user-end ID tables differ between institutions and hospitals, and the device codes for different devices are also different.

[0034] Furthermore, the data acquisition subsystem 3 includes: a data acquisition device and an intelligent suction device; the data acquisition device includes at least: a pressure sensor, a miniature camera, a flow sensor, a sound sensor, a body temperature sensor, and a breathing sensor band.

[0035] The pressure sensor is built into the smart suction device to collect pressure data.

[0036] A miniature camera is installed at the end of the smart suction device or at the sputum collection device to collect images of sputum.

[0037] A flow sensor is installed in the suction tube of the smart suction device to collect sputum flow rate.

[0038] The sound sensor is placed near the patient's airway or on the suction catheter to collect respiratory sound data.

[0039] The suctioning frequency is obtained by recording the switching actions (e.g., the number of times the switch is turned on and the time interval between the switching on and off) by the control system or statistical equipment of the intelligent suction device.

[0040] The body temperature sensor is used to collect the patient's body temperature.

[0041] The breathing sensor band is used to detect the amplitude and frequency of chest or abdominal movements in patients to obtain their respiratory rate.

[0042] Furthermore, such as Figure 2As shown in the illustration, in one embodiment, the collection subsystem 3 further includes a closed suction device system 31. The closed suction device system 31 includes a lever-type negative pressure switch 311, a suction tube 312, a suction tube connector 313, a closed suction tube three-way assembly 314, and an endotracheal tube 315. One end of the lever-type negative pressure switch 311 is connected to one end of the suction tube 312, and the other end of the lever-type negative pressure switch 311 is a negative pressure end; the other end of the suction tube 312 is connected to one end of the suction tube connector 313, and the other end of the suction tube connector 313 is connected to one end of the closed suction tube three-way assembly 314; the other end of the closed suction tube three-way assembly 314 is connected to one end of the endotracheal tube 315. The suction tube 312 is covered with a membrane. The sealed suction catheter three-way assembly 314 has a built-in suction guide. The diameter of the end of the suction guide facing the suction device connector 313 is smaller than the diameter of the other end of the suction guide facing the endotracheal tube 315 (i.e., the inner hole of the suction guide has a preset taper). When the lever-type negative pressure switch 311 is in the open state, the lever-type negative pressure switch 311, the suction catheter 312, the suction catheter connector 313, the suction guide of the sealed suction catheter three-way assembly 314 and the endotracheal tube 315 are connected.

[0043] Specifically, a suction guide is integrated into the closed suction tube tee assembly 314 of the closed suction device system 31. This guide allows the suction tube 312 to be pushed forward in a straight line through its tapered inner hole, solving the impact problem of traditional flat connectors. The suction guide directly matches the tapered outer connector of the endotracheal tube 315 without the need for bolts or clips, ensuring a tight seal, simplifying the assembly process, and reducing production costs.

[0044] like Figure 3 As shown, this application provides a data analysis and early warning method for intelligent suction devices, including the following steps:

[0045] S1: Receive the collection request sent by the user terminal, determine the collection execution device from the corresponding collection subsystem according to the collection request, and establish a communication channel with the collection execution device; wherein, the collection request includes at least: patient medical data, initial vital signs data and user terminal ID.

[0046] Specifically, the initial vital signs data are: the patient's vital signs data collected in real time when the user uploads the collection request. The initial vital signs data includes a variety of vital signs indicators, including at least: respiratory function indicators, as well as circulatory and baseline status indicators.

[0047] The respiratory function indicators include at least: oxygen saturation (SpO2), respiratory rate (RR), tidal volume, and inspiratory negative pressure. Tidal volume is applicable to mechanically ventilated patients. Inspiratory negative pressure reflects the ability to spontaneously clear sputum.

[0048] Circulatory and basal status indicators include at least: heart rate (HR), blood pressure (BP), and body temperature. Among these, body temperature is used to assess infectious activity.

[0049] Furthermore, the sub-steps for determining the acquisition execution device from the corresponding acquisition subsystem based on the acquisition request and establishing a communication channel with the acquisition execution device are as follows:

[0050] S11: Based on the collection request, traverse and analyze the pre-built device database to obtain device recommendation information, and send the device recommendation information to the user terminal; wherein, the device recommendation information includes at least: recommended device code and recommended device parameters.

[0051] Furthermore, the sub-steps for traversing and analyzing the pre-built device database based on the collection request to obtain device recommendation information are as follows:

[0052] S111: Traverse the device database based on the user ID, and take the sub-device database corresponding to the user ID table to which the user ID belongs as the target sub-device database.

[0053] S112: Extract features from patient medical data to obtain population attribute features, determine the current population attribute based on the population attribute features, and use the current population attribute to traverse each device data packet in the target sub-device database, and take the device data packet whose standard population attribute matches the current population attribute as the target device data packet.

[0054] Specifically, population attribute characteristics are features in patient medical data that can accurately represent population attributes. Population attributes include at least: adults, children, the elderly, premature infants, infants, patients with respiratory infections, patients with pneumonia, stroke patients, patients with neurogenic dysphagia, and patients with tracheotomy.

[0055] S113: Extract features from patient medical data to obtain current symptom features, calculate the type attribution value between the standard symptom feature data packet corresponding to each sub-device data packet in the target device data packet and the current symptom features, determine the sub-device data packet corresponding to the maximum type attribution value as the target sub-device data packet, and use the device code corresponding to the target sub-device data packet as the recommended device code.

[0056] Specifically, the current symptom features are those features in the patient's medical data that can accurately characterize the symptom type. The larger the type attribution value, the more consistent the current symptom feature is with the symptom type represented by the standard symptom feature data package.

[0057] S114: Extract features from patient medical data to obtain disease severity features, calculate the degree attribution value between the disease severity features and each standard disease severity feature data packet in the standard disease severity table corresponding to the target sub-device data packet, and take the disease severity value corresponding to the maximum degree attribution value as the initial disease severity value.

[0058] Specifically, disease severity features are features in a patient's medical data that can accurately characterize the severity of the disease. Patients' medical data (such as electronic medical records, examination reports, test results, etc.) contain multiple disease severity features that can directly or indirectly reflect the severity of the disease. Disease severity features can be extracted through structured analysis.

[0059] S115: Analyze the initial vital signs data to obtain the initial vital signs abnormality value.

[0060] Furthermore, the expression for the initial abnormality level of vital signs is:

[0061]

[0062] Where Ytz is the initial abnormality value; Zb i Zb represents the abnormal value of the i-th vital sign indicator in the initial vital sign data. When the deviation of the i-th vital sign indicator from the preset normal reference range is within the normal range, Zb... i =A1, when the deviation of the i-th vital sign index from the preset normal reference range is slightly abnormal, Zb i =A2, when the deviation of the i-th vital sign index from the preset normal reference range is significantly abnormal, Zb i =A3, A1, A2, and A3 are all natural numbers, A1 < A2 < A3; i ∈ [1, I], where I is the total number of types of vital signs in the initial vital signs data.

[0063] Specifically, the vital signs indicators include at least: respiratory function indicators, as well as circulatory and baseline status indicators. Both respiratory function indicators and circulatory and baseline status indicators have clearly defined normal reference ranges. Abnormal values ​​are categorized into three levels—normal, mildly abnormal, and significantly abnormal—based on the extent to which these indicators deviate from the normal reference range. The specific values ​​for A1, A2, and A3 are set according to the actual situation; the greater the degree of abnormality, the larger the abnormal value.

[0064] S116: Calculate the comprehensive severity value based on the initial symptom severity value and the initial abnormality value of vital signs. Use the comprehensive severity value to traverse the standard severity range in the target sub-device data packet and use the standard device parameters corresponding to the standard severity range to which the comprehensive severity value belongs as the recommended device parameters.

[0065] Furthermore, the expression for the overall degree value is:

[0066] Yzh = μ1·Ybq + μ2·Ytz;

[0067] Where Yzh is the overall severity value; Ybq is the initial symptom severity value; Ytz is the initial sign abnormality value; μ1 is the weight of the initial symptom severity; μ2 is the weight of the initial sign abnormality, and μ1+μ2=1.

[0068] Specifically, the values ​​of μ1 and μ2 are determined based on practical experience and research data.

[0069] S117: Use the recommended device code and recommended device parameters as device recommendation information.

[0070] Specifically, the equipment recommendation information obtained through analysis in this application can provide medical staff with reference information regarding equipment models and parameter settings. This reference information can improve the efficiency and rationality of medical staff's decision-making, while taking into account both professionalism and safety. Specifically, this includes, but is not limited to, using this equipment recommendation information to filter out solutions that better suit the needs of the scenario from numerous options, reducing the time cost for medical staff to compare and select options one by one. This equipment recommendation information combines factors such as equipment performance and clinical scenario adaptability, helping medical staff to more accurately match equipment models (e.g., selecting suction devices with appropriate functions for different patients) and providing professional guidance for parameter settings (e.g., suction power, frequency, etc.), reducing selection bias caused by insufficient experience or incomplete information. This equipment recommendation information considers standard operating procedures or past effective cases, providing medical staff with a basic framework for safe operation. Adjustments can then be made based on individual patient differences, making precise settings easier to achieve.

[0071] S12: After receiving the user terminal's reference device recommendation information, send the device information to be used. The device information to be used includes at least: device code, multiple parameter configuration items, and parameter values ​​corresponding to each parameter configuration item.

[0072] Specifically, the parameter configuration items include at least: negative pressure parameters, suctioning frequency, suctioning time and interval, humidification parameters, and sputum image acquisition frequency.

[0073] The user terminal configures the device corresponding to the device code according to the parameter values ​​of the parameter configuration items or the custom parameter values, and sends the device usage information after the configuration is completed.

[0074] S13: Determine that the device corresponding to the device code is the acquisition and execution device, and establish a communication channel with the acquisition and execution device.

[0075] Yes, the analysis and early warning center and the data acquisition and execution equipment establish a communication channel through a wireless connection.

[0076] S2: Receive the raw acquisition data sent by the acquisition execution device through the communication channel, process the raw acquisition data to obtain denoised data and normalized data, and use the raw acquisition data, denoised data and normalized data as multi-source data; wherein, the raw acquisition data includes at least: pressure data, sputum image, sputum flow, respiratory sound data, suctioning frequency, patient body temperature and patient respiratory rate.

[0077] Pressure data refers to the negative or positive pressure data within the tubing during the operation of the suction device, reflecting the pressure fluctuations during device operation. Pressure data can be collected using pressure sensors.

[0078] The noise-reduced data should include at least: noise-reduced pressure data, noise-reduced sputum flow rate, and noise-reduced respiratory sound data.

[0079] The normalized data should include at least the normalized pressure data and the normalized sputum flow rate.

[0080] Further, the sub-steps for processing the original collected data to obtain denoised and normalized data are as follows:

[0081] S21: Wavelet transform algorithm is used to denoise the pressure data to obtain denoised pressure data; normalization is performed on the denoised pressure data to obtain normalized pressure data.

[0082] S22: The sputum flow rate is denoised using the Kalman filter method to obtain the denoised sputum flow rate; the denoised sputum flow rate is then normalized to obtain the normalized sputum flow rate.

[0083] S23: Use a filtering algorithm to denoise the breathing sound data to obtain denoised breathing sound data.

[0084] S3: Perform multi-level analysis on multi-source data to obtain multi-level early warning results, and send the multi-level early warning results to the user terminal. The multi-level early warning results include at least: real-time early warning results and predictive early warning results.

[0085] Among them, real-time early warning results include at least: real-time equipment early warning results and real-time health early warning results; predictive early warning results include at least: pressure prediction risk results, flow prediction risk results and health prediction risk results.

[0086] Specifically, the multi-level early warning results provide medical staff with a basis for adjusting equipment parameters, which can help them to more comprehensively and accurately judge equipment risks and patient health risks, and adjust operations in a timely manner.

[0087] Furthermore, the sub-steps for performing multi-level analysis on multi-source data to obtain multi-level early warning results are as follows:

[0088] S31: Perform real-time analysis of multi-source data to obtain real-time early warning results.

[0089] Specifically, real-time early warning results include at least: real-time equipment early warning results and real-time health early warning results.

[0090] Furthermore, the sub-steps for performing real-time analysis of multi-source data to obtain real-time early warning results are as follows:

[0091] S311: Perform equipment early warning analysis based on the denoised data from multi-source data to generate real-time equipment early warning results, where the real-time equipment early warning results are either risky or risk-free.

[0092] Furthermore, the sub-steps for performing equipment early warning analysis based on the denoised data from the multi-source data to generate real-time equipment early warning results are as follows:

[0093] S3111: Calculate the pressure fluctuation entropy value based on the denoised pressure data in the denoised data, use the pressure fluctuation threshold to judge the pressure fluctuation entropy value, and generate a real-time warning result for the equipment. If the pressure fluctuation entropy value is greater than the pressure fluctuation threshold, it is determined that there is a risk of pipeline blockage, and the generated real-time warning result for the equipment is risky; if the pressure fluctuation entropy value is less than or equal to the pressure fluctuation threshold, then execute S3113.

[0094] Specifically, the specific value of the pressure fluctuation threshold is set based on actual experience or research data.

[0095] Furthermore, the expression for the pressure fluctuation entropy value is:

[0096]

[0097] Where Bds is the pressure fluctuation entropy value, used to measure the degree of fluctuation in the denoised pressure data; P n The probability of the nth pressure fluctuation state occurring; log(P) n ) represents the logarithm of the probability of the nth pressure fluctuation state; n∈[1,N], where N is the total number of pressure fluctuation state categories in the denoised pressure data.

[0098] Specifically, due to P n The normalized probability value, 0 < P n <1, log(P) n ) is a negative value, and Bds is a positive value.

[0099] This application quantifies blockage risk by using pressure fluctuation entropy values, which is more accurate than traditional pressure data analysis methods (i.e., setting a fixed traditional threshold, collecting pressure data, and directly comparing whether the pressure data exceeds the traditional threshold).

[0100] S3112: Calculate the flow fluctuation entropy value based on the denoised sputum flow rate in the denoised data, use the flow fluctuation threshold to judge the flow fluctuation entropy value, and generate a real-time warning result for the equipment. If the flow fluctuation entropy value is greater than the flow fluctuation threshold, it is determined that the equipment has poor operating stability, and the generated real-time warning result for the equipment is risky; if the flow fluctuation entropy value is less than or equal to the flow fluctuation threshold, then execute S3113.

[0101] Specifically, the specific value of the traffic fluctuation threshold is set based on actual experience or research data.

[0102] Furthermore, the expression for the entropy value of flow fluctuation is:

[0103]

[0104] Where Bll is the flow fluctuation entropy value, which is an indicator of the degree of disorder in the sequence fluctuation of denoised sputum flow; q j Let be the probability distribution value of the j-th type of traffic sample; log(q) j ) for q j Take the logarithm; j∈[1,J], where J is the total number of samples of denoised sputum flow.

[0105] Specifically, q j The q values ​​were obtained by normalizing the denoised sputum flow rate. j The sum of q equals 1, used to reflect q j The proportion in the overall traffic sequence. log(q) j ) is used to convert probability values ​​into units of information measurement. Because q j The normalized probability value, 0 < q j <1, log(q) j ) is a negative value, and Bll is a positive value.

[0106] S3113: Set the analysis period. During the analysis period, extract the local maximum value from the denoised pressure data as the pressure peak value, and extract the local minimum value from the denoised sputum flow rate as the flow trough value. Use the difference between the pure value of the pressure peak value and the pure value of the flow trough value as the matching degree of the peak and trough values. Use the preset matching degree threshold to judge the matching degree of the peak and trough values ​​and generate a real-time warning result for the device. If the matching degree of the peak and trough values ​​is greater than the preset matching degree threshold, it is determined that the suction pump power is abnormal and the generated real-time warning result for the device is risky. If the matching degree of the peak and trough values ​​is less than or equal to the preset matching degree threshold, the generated real-time warning result for the device is risk-free.

[0107] Specifically, the specific method for setting the analysis period depends on the actual situation. This application preferably uses a rolling time period, which is [previous endpoint time, current time]. There is a preset interval between the current time and the previous endpoint time, and the specific duration of the preset interval is set according to the actual situation. The current time is the time of the original data collection for this analysis. The rolling time period ensures that each analysis includes the latest real-time data, improving the accuracy and timeliness of the analysis results. The pure numerical values ​​of pressure peaks and flow valleys are values ​​that ignore physical units, i.e., the influence of units is removed. The specific value of the preset matching degree threshold is set based on actual experience or research data.

[0108] S312: Perform health early warning analysis based on the denoised data and the original collected data from the multi-source data to generate real-time health early warning results, where the real-time health early warning results are either risky or risk-free.

[0109] Furthermore, the sub-steps for generating real-time health early warning results by performing health early warning analysis based on the denoised data and the original collected data from the multi-source data are as follows:

[0110] S3121: Calculate respiratory health risk values ​​based on a multifactor linear regression model.

[0111] Furthermore, the expression for respiratory health risk value is:

[0112] Fjk=a·σ p +b·Pxtp+c·Xtx+d·Twd+e·Phxp+f·Ftx+g·Ssy+h;

[0113] Where Fjk is the respiratory health risk value; σ p The standard deviation of pressure fluctuation is calculated using the denoised pressure data; Pxtp is the suctioning frequency; Xtx is the similarity value between the color feature vector and the infection feature vector of the sputum image; Twd is the patient's body temperature; Phxp is the patient's respiratory rate; Ftx is the feature parameter of the sputum image; Ssy is the respiratory audio analysis parameter; a, b, c, d, e, f, g, and h are regression coefficients.

[0114] Specifically, σ p Used to reflect the stability of equipment pressure. Suctioning frequency Pxtp is the ratio of the number of suctionings per unit time to the unit time. Xtx is used to reflect the risk of respiratory tract infection. Phxp is the number of waveform cycles per unit time (preferably within 1 minute) calculated by analyzing periodic signals collected by devices such as chest and abdominal impedance sensors, respiratory flow sensors, or accelerometers.

[0115] The optimal values ​​of regression coefficients a, b, c, d, e, f, g, and h can be obtained by fitting the model to the training data. The training data includes input feature data and label data (actual risk). Input feature data includes at least: historical pressure fluctuation standard deviation, historical suctioning frequency, similarity value between the color feature vector and infection feature vector of historical sputum images, historical patient body temperature, historical patient respiratory rate, feature parameters of historical sputum images, and historical respiratory audio analysis parameters. Label data (actual risk) is model learning data defined based on clinical diagnostic results. The model can be Ordinary Least Squares (OLS), Ridge Regression, Lasso Regression, or Elastic Net, but is not limited to these methods.

[0116] Furthermore, the collected sputum images are preprocessed to obtain preprocessed sputum images. The preprocessing includes at least denoising, cropping, and preserving effective regions. The RGB color space of the preprocessed sputum images is converted to the HSV color space, which is more consistent with human visual perception (where H is the hue, S is the saturation, and V is the brightness), reducing illumination interference. Histogram features (such as bin count, peak position, and distribution variance) of each channel (e.g., the H channel) are extracted, or color center values ​​are obtained through pixel clustering. The histogram features or color center values ​​are arranged in a preset order to form a color feature vector (i.e., a multidimensional vector) of the sputum image, which is used to calculate similarity with the infection feature vector.

[0117] The patient's body temperature is collected using an infrared temperature sensor, a contact thermocouple sensor, or an electronic thermometer, but not limited to these methods.

[0118] Ftx represents the sputum viscosity feature value, the purulent component ratio feature value, or the degree of sputum foaming. Feature parameters of the sputum image are obtained based on its morphology and composition characteristics. Specifically, the sputum viscosity feature value is calculated using the sputum's fluidity (e.g., edge diffusion speed) and texture roughness (e.g., gray-level co-occurrence matrix). The Otsu algorithm is used to perform binarization segmentation on the sputum image, distinguishing purulent regions (e.g., high-brightness, specific color regions) from non-purulent regions. The proportion of pixels in purulent regions to the total effective region is calculated as the purulent component ratio feature value. Quantitative parameters corresponding to the degree of sputum foaming (e.g., bubble number / area ratio) and color intensity (e.g., mean gray value) are extracted based on key clinical indicators.

[0119] Ssy is the frequency characteristic value of breath sounds or the intensity characteristic value of abnormal breathing sounds, obtained based on the collected breath audio data. Specifically, the breath sound data (e.g., inhalation / exhalation sounds collected by a microphone) is preprocessed (e.g., denoising, extracting effective segments) to obtain denoised breath sound data; the time-domain signal of the denoised breath sound data is converted to the frequency domain through Fourier transform to extract frequency features (e.g., dominant frequency, proportion of high-frequency components); the intensity (sound pressure level) and duration of abnormal breath sounds (e.g., rales, wheezing) are analyzed, or the texture features of the audio are extracted through Mel-frequency cepstral coefficients (MFCC); key features (e.g., abnormal sound intensity, dominant frequency value) are selected as breath audio analysis parameters.

[0120] S3122: Analyze respiratory health risk values ​​using preset respiratory health risk thresholds and generate real-time health warning results. If the respiratory health risk value is greater than the preset respiratory health risk threshold, the generated real-time health warning result is "at risk"; if the respiratory health risk value is less than or equal to the preset respiratory health risk threshold, the generated real-time health warning result is "no risk".

[0121] Specifically, the specific values ​​for respiratory health risk thresholds are set based on practical experience and research results.

[0122] S32: Perform predictive analysis on multi-source data to obtain predictive and early warning results.

[0123] Specifically, the forecast and early warning results should include at least: pressure forecast risk results, flow forecast risk results, and health forecast risk results.

[0124] Furthermore, the sub-steps for performing predictive analysis on multi-source data to obtain predictive and early warning results are as follows:

[0125] S321: Input the normalized stress data from the normalized data into the pre-trained stress prediction model. The stress prediction model outputs the stress prediction result. The stress prediction result is judged by the preset stress risk threshold, and a stress prediction risk result is generated. If the stress prediction result is greater than the preset stress risk threshold, the generated stress prediction risk result is risky; if the stress prediction result is less than or equal to the preset stress risk threshold, the generated stress prediction risk result is risk-free.

[0126] Furthermore, the pre-trained stress prediction model is a stress prediction model based on a bidirectional long short-term memory network (Bi-LSTM-Attention), but it is not limited to a stress prediction model based on a bidirectional long short-term memory network (Bi-LSTM-Attention). This application preferably uses a stress prediction model based on a bidirectional long short-term memory network (Bi-LSTM-Attention), which improves the accuracy of the prediction results.

[0127] Furthermore, the expression for the pressure prediction result is as follows:

[0128]

[0129] Where Yylsj is the pressure prediction result, σ ylsj is the activation function used for pressure prediction; This is the weight matrix of the fully connected layer in the pressure prediction model; This refers to the bias term of the fully connected layer in the pressure prediction model; This represents the hidden state in the stress prediction model after being weighted by the attention mechanism.

[0130] Specifically, Yylsj is the predicted result of pressure fluctuation trends. σ ylsj It is sigmoid, but not limited to sigmoid. These are the trainable parameters of the stress prediction model, obtained through backpropagation optimization using training data. During the training process of the stress prediction model, the parameters are continuously adjusted based on the error between the stress prediction results and the actual values ​​(e.g., the loss function). The value is calculated until the pressure prediction model converges. These are trainable parameters of the pressure prediction model, determined through backpropagation optimization of training data. They are used to adjust the output offset of the fully connected layer and improve the fitting ability of the flow prediction model. It can be calculated using the attention mechanism.

[0131] S322: Input the normalized sputum flow rate from the normalized data into the pre-trained flow prediction model. The flow prediction model outputs the flow prediction result. The flow prediction result is judged using a preset flow risk threshold to generate a flow prediction risk result. If the flow prediction result is greater than the preset flow risk threshold, the generated flow prediction risk result is risky; if the flow prediction result is less than or equal to the preset flow risk threshold, the generated flow prediction risk result is risk-free.

[0132] Furthermore, the pre-trained traffic prediction model is a traffic prediction model based on a bidirectional long short-term memory network (Bi-LSTM-Attention), but it is not limited to a traffic prediction model based on a bidirectional long short-term memory network (Bi-LSTM-Attention). This application preferably uses a traffic prediction model based on a bidirectional long short-term memory network (Bi-LSTM-Attention), which improves the accuracy of the prediction results.

[0133] Furthermore, the expression for the traffic prediction result is:

[0134]

[0135] Where Yflow is the traffic prediction result, σ flow is the activation function used for traffic prediction; This is the weight matrix of the fully connected layer in the traffic prediction model; This refers to the bias term of the fully connected layer in the traffic prediction model; This refers to the hidden state in the traffic prediction model after being weighted by the attention mechanism.

[0136] Specifically, Yflow is a prediction result of traffic fluctuation trends. σ flow It is sigmoid, but not limited to sigmoid. These are the trainable parameters of the traffic prediction model, obtained through backpropagation optimization using training data. During the training process of the traffic prediction model, the parameters are continuously adjusted based on the error between the traffic prediction results and the actual values ​​(e.g., the loss function). The value is calculated until the traffic prediction model converges. These are the trainable parameters of the traffic prediction model, determined through backpropagation optimization of the training data. They are used to adjust the output offset of the fully connected layer and improve the fitting ability of the traffic prediction model. It can be calculated using the attention mechanism.

[0137] S323: Input sputum image features, historical vital sign data, pressure prediction results, and flow prediction results into a pre-built health prediction model. The health prediction model outputs health prediction results. Use a preset health risk threshold to judge the health prediction results and generate a health prediction risk result. If the health prediction result is greater than the preset health risk threshold, the generated health prediction risk result is risky; if the health prediction result is less than or equal to the preset health risk threshold, the generated health prediction risk result is risk-free.

[0138] Specifically, historical vital signs data are data that can reflect the patient's overall physical function and health status. Historical vital signs data include at least: heart rate, blood pressure, and body temperature.

[0139] Furthermore, the pre-constructed health prediction model is a health prediction model based on support vector machines and generative adversarial networks, but it is not limited to a health prediction model based on support vector machines and generative adversarial networks. This application preferably uses a health prediction model based on support vector machines and generative adversarial networks, which improves the accuracy of the prediction results.

[0140] The data analysis and early warning method and system for intelligent suction devices proposed in this application solves the problems of incomplete data collection, imprecise processing, and inaccurate early warning in existing technologies. It can provide medical staff with more reliable equipment status monitoring and patient health risk prediction, assisting medical staff to take timely intervention measures, improving the safety and effectiveness of suction care, reducing medical risks, and optimizing the allocation of medical resources.

[0141] Although preferred embodiments of this application 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 scope of protection of this application is intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application. Obviously, those skilled in the art can make various alterations and variations to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of protection of this application and its equivalents, this application also intends to include these modifications and variations.

Claims

1. A data analysis and early warning method for a smart sputum suction device, characterized in that, The method comprises the following steps: S1: receiving a collection request sent by a user terminal, determining a collection execution device from a corresponding collection subsystem according to the collection request, and establishing a communication channel with the collection execution device; wherein the collection request at least includes patient medical data, initial vital sign data, and a user terminal ID; S2: receiving raw collection data sent by the collection execution device through the communication channel, processing the raw collection data to obtain denoised data and normalized data, and taking the raw collection data, the denoised data, and the normalized data as multi-source data; wherein the raw collection data at least includes pressure data, sputum image, sputum flow, respiratory sound data, sputum suction frequency, patient body temperature, and patient respiratory frequency; S3: performing multi-level analysis on the multi-source data to obtain multi-level early warning results, and sending the multi-level early warning results to the user terminal, wherein the multi-level early warning results at least include real-time early warning results and prediction early warning results.

2. The data analysis and early warning method for the intelligent sputum aspirator according to claim 1, characterized in that, The sub-steps of determining the collection execution device from the corresponding collection subsystem according to the collection request and establishing a communication channel with the collection execution device are as follows: S11: traversing and analyzing a pre-constructed device database according to the collection request to obtain device recommendation information, and sending the device recommendation information to the user terminal; wherein the device recommendation information at least includes recommended device codes and recommended device parameters; S12: receiving usage device information sent by the user terminal after referring to the device recommendation information, wherein the usage device information at least includes device codes, multiple parameter configuration items, and parameter values corresponding to each parameter configuration item; S13: determining the device corresponding to the device code as the collection execution device, and establishing a communication channel with the collection execution device.

3. The data analysis and early warning method for intelligent sputum suction device according to claim 1, characterized in that, The sub-steps of processing the raw collection data to obtain denoised data and normalized data are as follows: S21: using a wavelet transform algorithm to perform denoising processing on the pressure data to obtain denoised pressure data; and performing normalization processing on the denoised pressure data to obtain normalized pressure data; S22: using a Kalman filter method to perform denoising processing on the sputum flow to obtain denoised sputum flow; and performing normalization processing on the denoised sputum flow to obtain normalized sputum flow; S23: using a filtering algorithm to perform denoising processing on the respiratory sound data to obtain denoised respiratory sound data.

4. The data analysis and early warning method for the intelligent sputum aspirator according to claim 1, characterized in that, The sub-steps of performing multi-level analysis on the multi-source data to obtain multi-level early warning results are as follows: S31: performing real-time analysis on the multi-source data to obtain real-time early warning results, wherein the real-time early warning results at least include device real-time early warning results and health real-time early warning results; S32: performing prediction analysis on the multi-source data to obtain prediction early warning results, wherein the prediction early warning results at least include pressure prediction risk results, flow prediction risk results, and health prediction risk results.

5. The data analysis and early warning method for the intelligent sputum aspirator according to claim 4, characterized in that, The sub-steps of performing real-time analysis on the multi-source data to obtain real-time early warning results are as follows: S311: performing device early warning analysis according to the denoised data in the multi-source data to generate device real-time early warning results, wherein the device real-time early warning results are risk or no risk; S312: Perform health early warning analysis according to the denoised data and the original collected data in the multi-source data, and generate a health real-time early warning result, wherein the health real-time early warning result is at risk or not at risk.

6. The data analysis and early warning method for the intelligent sputum aspirator according to claim 5, characterized in that, The sub-steps of performing device early warning analysis according to the denoised data in the multi-source data to generate a device real-time early warning result are as follows: S3111: Calculate a pressure fluctuation entropy value based on the denoised pressure data in the denoised data, and use a pressure fluctuation threshold to judge the pressure fluctuation entropy value to generate a device real-time early warning result; if the pressure fluctuation entropy value is greater than the pressure fluctuation threshold, it is determined that there is a risk of pipeline blockage, and the generated device real-time early warning result is at risk; If the pressure fluctuation entropy value is less than or equal to the pressure fluctuation threshold, perform S3113; S3112: Calculate a flow fluctuation entropy value based on the denoised sputum flow in the denoised data, and use a flow fluctuation threshold to judge the flow fluctuation entropy value to generate a device real-time early warning result; if the flow fluctuation entropy value is greater than the flow fluctuation threshold, it is determined that the device is not stable, and the generated device real-time early warning result is at risk; If the flow fluctuation entropy value is less than or equal to the flow fluctuation threshold, perform S3113; S3113: Set an analysis period, extract a local maximum value from the denoised pressure data as a pressure peak value and extract a local minimum value from the denoised sputum flow as a flow valley value within the analysis period; take the difference between the pure value of the pressure peak value and the pure value of the flow valley value as the matching degree of the peak value and the valley value, and use a preset matching degree threshold to judge the matching degree of the peak value and the valley value to generate a device real-time early warning result; if the matching degree of the peak value and the valley value is greater than the preset matching degree threshold, it is determined that the sputum pump power is abnormal, and the generated device real-time early warning result is at risk; if the matching degree of the peak value and the valley value is less than or equal to the preset matching degree threshold, the generated device real-time early warning result is not at risk.

7. The data analysis and early warning method for the intelligent sputum aspirator according to claim 5, characterized in that, The sub-steps of performing health early warning analysis according to the denoised data and the original collected data in the multi-source data to generate a health real-time early warning result are as follows: S3121: Calculate a respiratory tract health risk value based on a multi-factor linear regression model; S3122: Use a preset respiratory tract health risk threshold to analyze the respiratory tract health risk value to generate a health real-time early warning result; if the respiratory tract health risk value is greater than the preset respiratory tract health risk threshold, the generated health real-time early warning result is at risk; If the respiratory tract health risk value is less than or equal to the preset respiratory tract health risk threshold, the generated health real-time early warning result is not at risk.

8. The data analysis and early warning method for the intelligent sputum aspirator according to claim 4, characterized in that, The sub-steps of performing prediction analysis on the multi-source data to obtain a prediction early warning result are as follows: S321: Input the normalized pressure data in the normalized data into a pre-trained pressure prediction model, output a pressure prediction result from the pressure prediction model, use a preset pressure risk threshold to judge the pressure prediction result, and generate a pressure prediction risk result; if the pressure prediction result is greater than the preset pressure risk threshold, the generated pressure prediction risk result is at risk; if the pressure prediction result is less than or equal to the preset pressure risk threshold, the generated pressure prediction risk result is not at risk; S322: input the normalized sputum flow in the normalized data into the pre-trained flow prediction model, output the flow prediction result from the flow prediction model, judge the flow prediction result by using the preset flow risk threshold, generate the flow prediction risk result, if the flow prediction result is greater than the preset flow risk threshold, the generated flow prediction risk result is risky; if the flow prediction result is less than or equal to the preset flow risk threshold, the generated flow prediction risk result is risk-free; S323: input the sputum image feature, the historical vital sign data, the pressure prediction result and the flow prediction result into the pre-constructed health prediction model, output the health prediction result from the health prediction model, judge the health prediction result by using the preset health risk threshold, generate the health prediction risk result, if the health prediction result is greater than the preset health risk threshold, the generated health prediction risk result is risky; if the health prediction result is less than or equal to the preset health risk threshold, the generated health prediction risk result is risk-free.

9. The data analysis and early warning method for the intelligent sputum aspirator according to claim 8, characterized in that, The pre-trained pressure prediction model is a pressure prediction model based on a bidirectional long short-term memory network; the pre-trained flow prediction model is a flow prediction model based on a bidirectional long short-term memory network; and the pre-constructed health prediction model is a health prediction model based on a support vector machine and a generative adversarial network.

10. A data analysis and early warning system for a smart sputum suction device, characterized in that, It comprises: an analysis and early warning center and at least one user terminal, each user terminal corresponding to at least one acquisition subsystem; wherein the analysis and early warning center is used to execute the data analysis and early warning method for the intelligent sputum suction device in any one of claims 1-9; the user terminal: for sending acquisition requests and using device information; for receiving multi-level early warning results; the acquisition subsystem: supporting the analysis and early warning center and the selected acquisition execution device to establish a communication channel, and the acquisition execution device sending original acquisition data to the analysis and early warning center through the communication channel.

Images