A rapid calibration system and method for an ultrasonic lung function test device

Abstract

The application provides a rapid calibration system and method of an ultrasonic lung function detection device, which is used to solve the problem that the efficiency, accuracy and consistency of the existing calibration scheme are difficult to balance. The method steps are as follows: using a mounting jig to install the device to be calibrated, and verifying the reference distance of the two pressure races; connecting a standard flow generator to collect single-point data at the characteristic flow point; using the offset method to correct the reference mean FVC correction table and the collected single-point offset to synthesize the complete FVC correction table at any temperature; inputting the FVC characteristics into the pre-trained SVR model to output the complete PEF correction table; and burning the FVC correction table and the PEF correction table to the device. The application unifies the mechanical reference of all devices to be calibrated from the physical layer by using the mounting jig, and eliminates the largest systematic variation source; using the data-driven intelligent algorithm, based on a few measured data points, the complete high-precision correction table of the device is quickly inferred and generated, and the efficiency and accuracy of the correction are improved.

Description

Technical Field

[0001] This invention relates to the field of ultrasound equipment calibration, and in particular to a rapid calibration system and method for an ultrasound pulmonary function testing device. Background Technology

[0002] As a key device for the diagnosis and monitoring of chronic respiratory diseases (such as asthma and chronic obstructive pulmonary disease), the accuracy of ultrasound pulmonary function testing directly affects the reliability of clinical diagnosis. To ensure measurement accuracy, each device undergoes rigorous nonlinear calibration of its core indicators—forced vital capacity (FVC) and peak expiratory flow (PEF)—before leaving the factory, generating a device-specific flow-calibration lookup table (LUT). The currently prevalent calibration method in the industry is the full flow point scanning calibration method. This method requires the use of a high-precision standard flow generator (such as a piston-type respiratory simulator) to perform ventilation tests at dozens of preset flow points. By comparing the device's original readings with the standard values, the calibration coefficient for each point is calculated, ultimately fitting a complete calibration curve. This process essentially establishes a unique "acoustic-system response" characteristic model for each device.

[0003] However, this method has the following inherent defects that are difficult to overcome: (1) Efficiency bottleneck: The calibration process is extremely cumbersome and time-consuming. Taking a disposable pipeline as an example, a single device needs to conduct tests at nearly 30 flow points to complete the full calibration of the two indicators FVC and PEF. Each point usually needs to be repeated multiple times to take the average to ensure stability, with a total time consumption of 70-100 minutes. This has become the core bottleneck restricting the throughput and capacity of the production line. (2) Consistency dilemma ("one machine, one meter"): Due to the unavoidable slight differences in the sensitivity, installation angle, coupling state and hardware circuit component parameters of the ultrasonic transducer, coupled with the influence of ambient temperature fluctuations, the system response functions of different devices, and even the same device, are different under different conditions. Therefore, the calibration table generated by the above method has a strong "device-specific" characteristic and cannot be used universally between different devices. This forces the production end to perform a complete independent calibration for each device, making it impossible to achieve rapid batch calibration, which significantly increases production costs. (3) High dependence on operator skills: The whole process relies heavily on skilled technicians to connect equipment, set parameters, record data and perform manual calculations. The degree of automation is low, the risk of human error is high, and it is not conducive to standardized quality control.

[0004] To address these issues, existing research focuses on two main directions: 1. Hardware consistency improvement: This involves reducing inherent differences between devices by precisely selecting sensors and electronic components and optimizing mechanical structure design tolerances. However, this method is costly and cannot completely eliminate random errors introduced by assembly processes. For ultrasonic measurement systems, which are highly sensitive to installation, its improvement effect is limited. 2. Software algorithm compensation: This involves using mathematical methods such as linear regression, polynomial fitting, or lookup table interpolation to post-process and compensate for measurement data. While these methods can correct systematic errors to some extent, they often rely on a large amount of prior calibration data. Furthermore, for indicators like PEF, which are strongly correlated with dynamic response and time series, the compensation capability of traditional static models is insufficient, making it difficult to fundamentally improve calibration efficiency.

[0005] In summary, existing technical solutions either fail to address efficiency bottlenecks or struggle to balance accuracy and consistency, and a comprehensive technical solution capable of systematically and fundamentally overcoming the contradiction between production efficiency and quality consistency in ultrasonic pulmonary function testing has yet to be developed. Therefore, there is an urgent need for an innovative calibration system and method to achieve an order-of-magnitude increase in production efficiency and a high degree of consistency in product parameters while ensuring clinical-grade measurement accuracy. Summary of the Invention

[0006] The purpose of this invention is to provide a rapid calibration system and method for ultrasound pulmonary function testing equipment. This addresses the technical challenge of balancing efficiency, accuracy, and consistency in existing calibration methods.

[0007] First, this application provides a rapid calibration system for an ultrasound pulmonary function testing device, including a mounting fixture for positioning and installing the device to be calibrated, the mounting fixture including a base;

[0008] A linear cylinder and a passive positioning plug are installed on the base. The telescopic end of the linear cylinder is connected to one end of the elastic push-pull module, and the other end of the elastic push-pull module is connected to the active positioning plug. When the calibration equipment is installed, the equipment component to be calibrated is placed between the passive positioning plug and the active positioning plug.

[0009] The elastic push-pull module includes a limiting block installed on the base, and the limiting block has a through hole arranged along the movement direction of the linear cylinder;

[0010] A sleeve is slidably disposed within the through hole, and a spring is installed inside the sleeve. One end of the sleeve is connected to the active positioning plug, and one end of the spring abuts against one end of the active positioning plug. The other end of the spring is connected to the front end of the floating joint rod. The front end of the floating joint rod can slide along the sleeve, and the rear end of the floating joint rod is connected to the telescopic end of the linear cylinder.

[0011] Optionally, the device to be calibrated includes a pipe housing and an ultrasonic transducer, wherein the ultrasonic transducer is pressed into the pipe housing by a mounting fixture.

[0012] Optional features also include a standard flow generator and processor;

[0013] During equipment calibration, the standard flow generator is connected to the device to be calibrated and provides a stable airflow to the device. The processor is used to calibrate the device based on the data collected by the flow sensor of the device to be calibrated.

[0014] Secondly, this application provides a rapid calibration method for an ultrasound pulmonary function testing device, employing the aforementioned rapid calibration system for ultrasound pulmonary function testing devices. The specific steps are as follows:

[0015] S1: Install the equipment to be calibrated using the mounting fixture and verify that the reference distance between the two pressure points is acceptable;

[0016] S2: Connect to a standard flow generator to collect single-point data at characteristic flow points;

[0017] S3: Using the offset method, a complete FVC correction table at any temperature is synthesized by combining the reference mean FVC correction table and the collected single-point offset.

[0018] S4: Input the FVC features into the pre-trained SVR model and output the complete PEF correction table;

[0019] S5: Burn the FVC calibration table and PEF calibration table to the device, verify the results, and the calibration is complete.

[0020] Optionally, the specific steps in step S1 are as follows:

[0021] S1.1: Place the pipe shell and ultrasonic transducer between the two passive positioning plugs and the active positioning plug to complete the initial positioning and acoustic axis alignment of the ultrasonic transducer;

[0022] S1.2: Start the linear cylinder. The linear cylinder pushes the elastic push-pull module and the ultrasonic transducer forward along the guide axis, and the ultrasonic transducer enters the pipe shell.

[0023] S1.3: The linear cylinder continues to advance, the spring is compressed, and a linearly increasing pressing force is generated, which smoothly and accurately presses the ultrasonic transducer into the designed position of the pipe shell;

[0024] S1.4: After the pressure is applied to the position, the linear cylinder air circuit briefly maintains pressure to ensure that the ultrasonic transducer is stably positioned, completing the first installation of the equipment to be calibrated.

[0025] Optionally, the specific steps in step S3 are as follows:

[0026] S3.1: Under production conditions with the application of mounting fixtures, collect FVC calibration tables for multiple devices at various ambient temperatures;

[0027] S3.2: Calculate the arithmetic mean of multiple FVC correction tables at each flow point to generate a reference mean FVC correction table decoupled from a specific temperature. ;

[0028] S3.3: Place the new device to be calibrated at the current ambient temperature. Next, select a specific flow point. Perform a ventilation test and compare the results to determine the operating conditions of the device to be calibrated at the current temperature. Below, at the flow point Measured correction coefficient at the location ;

[0029] S3.4: Calculate the translation of a single point for: The single-point translation amount Used to characterize the current calibrated equipment due to ambient temperature. The coefficient deviation relative to the average level caused by its own minor hardware deviations;

[0030] S3.5: Calculate the single-point offset Correction table with reference mean Simply add them together to generate the value of the device to be calibrated at the current temperature. Complete FVC correction table below for: .

[0031] Optionally, the specific steps in step S4 are as follows:

[0032] S4.1: Using the flow approximation criterion and rounding matching, m sets of flow point pairs with clear physical correspondence were established;

[0033] S4.2: Rapidly generate FVC features at the above m flow points using the offset method: correction coefficient vector ;

[0034] S4.3: Convert the correction coefficient vector Input the pre-trained SVR model and output the correction coefficient vector at m preset PEF flow points. Obtain the complete PEF calibration table.

[0035] Optionally, in step S4, m independent SVR models are trained for each of the m preset PEF flow points; the SVR model for the i-th preset PEF flow point is in the following form:

[0036] ;

[0037] in, The number of training samples. and It is a Lagrange multiplier and satisfies the constraints. , For penalty parameters, For bias terms, For radial basis function kernels, kernel function The radial basis function (RBF) kernel is used, specifically in the form of:

[0038] ;

[0039] in, The kernel function parameters control the degree of influence of a single training sample.

[0040] Because of the adoption of the above technical solution, the present invention has the following advantages:

[0041] 1. This application unifies the mechanical reference of all equipment to be calibrated at the physical level by using a high-precision installation fixture, eliminating the largest source of systematic variation; and utilizes a data-driven intelligent algorithm to quickly infer and generate a complete high-precision calibration table for the equipment based on a very small number of measured data points, thereby improving the efficiency and accuracy of calibration.

[0042] 2. The installation fixture in this application effectively unifies the hardware installation standards of the equipment, improves the consistency of the calibration table to an ideal level, and lays a solid foundation for the subsequent implementation of intelligent calibration schemes.

[0043] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0044] The accompanying drawings of this invention are described below.

[0045] Figure 1 This is a schematic diagram of the structure of the mounting fixture of the present invention.

[0046] Figure 2 This is a cross-sectional view of the mounting fixture for the present invention.

[0047] Figure 3 This is a flowchart of the rapid calibration method for the ultrasonic pulmonary function testing device of the present invention.

[0048] Figure 4 This is a flowchart illustrating the standardized installation process of the fixture for this invention.

[0049] Figure 5 This is a correction coefficient curve corresponding to the conventional installation method of the present invention.

[0050] Figure 6 This is a curve showing the correction coefficients corresponding to the standard installation fixture method of this invention.

[0051] Figure 7 This is the reference mean FVC correction plot for this invention.

[0052] Figure 8 This is a diagram illustrating the error characteristics of the offset method in this invention.

[0053] Figure 9 This is a comparison chart of the regression relationship between the predicted values ​​and the actual values ​​of different models in this invention.

[0054] In the diagram: 1-base; 2-linear cylinder; 3-passive positioning plug; 4-active positioning plug; 5-limiting block; 6-sleeve; 7-spring; 8-equipment to be calibrated; 801-pipe shell; 802-ultrasonic transducer; 9-floating joint rod. Detailed Implementation

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments. The terms "upper," "lower," "front," "rear," "left," "right," "vertical," and "horizontal," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the embodiments of the present invention. In the description of the embodiments of the present invention, it should be noted that, unless otherwise expressly specified and limited, the terms "connected" or "linked" should be interpreted broadly. For example, it can be a fixed connection, a detachable connection, an integral connection, an electrical connection, or a signal connection; it can be a direct connection or an indirect connection through an intermediate medium.

[0056] Example 1:

[0057] like Figures 1-2 The diagram illustrates a rapid calibration system for an ultrasonic pulmonary function testing device. The device to be calibrated, 8, includes a tubular housing 801 and an ultrasonic transducer 802, the transducer 802 being pressed into the tubular housing 801 using a mounting fixture. The rapid calibration system includes a standard flow generator and a processor; and a mounting fixture for positioning and installing the device to be calibrated, 8.

[0058] During equipment calibration, the standard flow generator is connected to the device to be calibrated 8 and provides a stable airflow to the device to be calibrated 8. The processor is used to calibrate the device to be calibrated based on the data collected by the flow sensor of the device to be calibrated 8.

[0059] The installation fixture includes a base 1; a linear cylinder 2 and a passive positioning plug 3 are installed on the base 1. The telescopic end of the linear cylinder 2 is connected to one end of an elastic push-pull module, and the other end of the elastic push-pull module is connected to an active positioning plug 4. When the calibration device 8 is installed, the components of the calibration device 8 are arranged between the passive positioning plug 3 and the active positioning plug 4.

[0060] The elastic push-pull module includes a limiting block 5 installed on the base 1. The limiting block 5 has a through hole arranged along the movement direction of the linear cylinder 2. A sleeve 6 is slidably arranged in the through hole. A spring 7 is installed in the sleeve 6. One end of the sleeve 6 is connected to the active positioning plug 4. One end of the spring 7 abuts against one end of the active positioning plug 4. The other end of the spring 7 is connected to the front end of the floating joint rod 9. The front end of the floating joint rod 9 can slide along the sleeve 6. The rear end of the floating joint rod 9 is connected to the telescopic end of the linear cylinder 2.

[0061] In this embodiment, the high-precision mounting fixture of this application unifies the mechanical reference of all devices to be calibrated at the physical level, eliminating the largest source of systematic variation. During calibration, the device to be calibrated is clamped and installed in the special mounting fixture. The fixture is pneumatically driven by the linear cylinder 2 and driven by the constant force of the elastic push-pull module to automatically complete the precise positioning and constant force pressing of the device to be calibrated (ultrasonic transducer). The high repeatability of the installation state is ensured by mechanical limit measurement verification.

[0062] As an embodiment of this application, in order to facilitate the placement of the pipe housing 801 and the ultrasonic transducer 802, a support seat (not shown in the figure) can be provided on the base 1 between the passive positioning plug 3 and the active positioning plug 4 for initial positioning of the pipe housing 801 and the ultrasonic transducer 802.

[0063] Experimental verification:

[0064] To evaluate the actual effect of the installation fixture designed in this application on improving the consistency of parameters between devices, this application designed a comparative experiment. Under the same baseline conditions, the parallel similarity between the calibration tables of multiple devices generated by the two modes of standardized installation using the fixture of this application and traditional manual installation was compared. This experiment collected calibration tables of 20 devices using the traditional manual installation method and 20 devices using the standardized installation method to compare the translational similarity index of the two cases.

[0065] When studying the effects of traditional installation methods and installation methods using standard mounting fixtures on calibration tables, the differences between the two installation methods can be observed simply by analyzing the translational similarity of the calibration tables for PEF and FVC. First, the calibration tables for FVC and PEF obtained from the traditional installation method and the installation method using standard mounting fixtures were processed to obtain the following results: Figure 5 and Figure 6 As shown, where Figure 5 (a) and Figure 5 (b) The correction coefficient curves for FVC and PEF obtained by the traditional installation method are shown below. Figure 6 (a) and Figure 6 (b) The correction coefficient curves of FVC and PEF obtained by the standard mounting fixture installation method of this application, respectively.

[0066] Depend on Figure 5 and Figure 6 The correction coefficient curves of PEF and FVC obtained by different installation methods are shown. From the comparison of the translational similarity of the correction coefficients obtained by the two installation methods, the translational similarity between the correction coefficient curves of PEF and FVC obtained by using the standard installation fixture method is better.

[0067] To scientifically evaluate the consistency of the calibration table between the two methods, this application employs a system analysis method based on translational similarity theory. This method, tailored to the characteristics of the calibration table in the ultrasound pulmonary ventilation function testing system, establishes a complete mathematical model and calculation process to evaluate the consistency differences between devices under different installation modes.

[0068] Let the set of devices include Units of equipment, each unit in There are correction coefficient measurements at the flow points, for the... The calibration coefficient sequence for this device is as follows:

[0069] (1)

[0070] For any two devices and Assuming there is a translation amount Make:

[0071] (2)

[0072] in, for A dimensional vector of all 1s This is the residual vector, which includes measurement noise and nonlinear differences.

[0073] The optimal translation is estimated using the least squares criterion:

[0074] (3)

[0075] Solving for the given information, we get:

[0076] (4)

[0077] in, and These are the arithmetic means of the two sequences, respectively.

[0078] The equipment after translation The data is as follows:

[0079] (5)

[0080] The correlation coefficient after translation is calculated as follows:

[0081] (6)

[0082] To comprehensively assess the consistency between devices, a systematic pairwise comparison method was adopted to construct a complete similarity matrix and extract statistical features.

[0083] Build Correlation coefficient matrix:

[0084] (7)

[0085] in, For equipment and The correlation coefficient matrix after optimal translation satisfies symmetry. .

[0086] Due to the autocorrelation coefficient The dataset lacks valid comparison information, and the matrix symmetry leads to information duplication. Extracting elements from the triangular matrix constructs a valid dataset.

[0087] (8)

[0088] Dataset Include Each independent comparison value, for In this situation, .

[0089] Based on dataset Calculate the average translational similarity:

[0090] (9)

[0091] The closer this value is to 1, the higher the consistency between devices.

[0092] The above data processing methods were used to process the correction tables of FVC and PEF obtained using traditional installation methods and standard installation fixtures to obtain the average correlation coefficient. The results are summarized in Table 1.

[0093] Table 1. Comparison of Average Translation Similarity of Correction Tables under Different Installation Methods

[0094]

[0095] As shown in Table 1, the consistency of calibration tables between devices was significantly improved after using the standard mounting fixture. The average translational similarity coefficients of FVC and PEF increased from 0.7947 and 0.8207 to 0.9989 and 0.9729, respectively, representing increases of 25.7% and 18.55%. The results indicate that the mounting fixture effectively unified the hardware installation benchmarks of the devices, improving the consistency of the calibration tables to an ideal level. This laid a solid foundation for the subsequent implementation of intelligent correction schemes.

[0096] Example 2:

[0097] like Figure 3 The rapid calibration method for an ultrasound pulmonary function testing device shown herein employs the rapid calibration system for the ultrasound pulmonary function testing device described in Example 1. The specific steps are as follows:

[0098] S1: Install the equipment to be calibrated using the mounting fixture and verify that the reference distance between the two pressure points is acceptable; if Figure 4 As shown, the specific steps are as follows:

[0099] S1.1: Place the pipe housing 801 and the ultrasonic transducer 802 between the two passive positioning plugs 3 and the active positioning plug 4 to complete the initial positioning and acoustic axis alignment of the ultrasonic transducer 802;

[0100] S1.2: Start linear cylinder 2. Linear cylinder 2 pushes the elastic push-pull module and ultrasonic transducer 802 forward along the guide axis. Ultrasonic transducer 802 enters pipe housing 801.

[0101] S1.3: As the linear cylinder 2 continues to advance, the spring 7 is compressed, generating a linearly increasing pressing force, which smoothly and accurately presses the ultrasonic transducer 802 into the designed position of the pipe shell 801.

[0102] S1.4: After the press-fit is in place, the air circuit of the linear cylinder 2 is briefly pressurized to ensure that the ultrasonic transducer 802 is stably in place, completing the first installation of the device 8 to be calibrated.

[0103] In this embodiment, the pressing force is determined by the fixed compression of spring 7 and remains constant. To ensure consistent installation, positional consistency is guaranteed by the fixed stroke of the cylinder and mechanical hard limit; force consistency is guaranteed by the spring 7. Guarantee, among which and All variables are controlled; angular consistency is ensured by the coaxiality of the elastic push-pull module axis and the pressure plug to ensure the alignment of the acoustic axis; for operational consistency, the entire process, except for the first step of placement, is automatically completed by the pneumatic system in a programmed manner, eliminating the variation caused by human intervention.

[0104] S2: Connect to a standard flow generator to collect single-point data at characteristic flow points;

[0105] S3: Using the offset method, a complete FVC correction table at any temperature is synthesized by combining the reference mean FVC correction table and the collected single-point offsets; the specific steps are as follows:

[0106] S3.1: Under production conditions with the application of mounting fixtures, collect FVC calibration tables for multiple devices at various ambient temperatures;

[0107] In this embodiment, the ambient temperature is 15℃~35℃, such as Figure 7 As shown, Figure 7 For reference mean FVC correction table. Figure 7 It characterizes the calibration properties of the equipment in an average sense and is incorporated into the production system as a benchmark.

[0108] S3.2: Calculate the arithmetic mean of multiple FVC correction tables at each flow point to generate a reference mean FVC correction table decoupled from a specific temperature. ;

[0109] S3.3: Place the new device to be calibrated at the current ambient temperature. Next, select a specific flow point. Perform a ventilation test and compare the results to determine the operating conditions of the device to be calibrated at the current temperature. Below, at the flow point Measured correction coefficient at the location ;

[0110] In this embodiment, a specific flow point The selected flow rate is -8L / s, as this flow rate point has stable airflow and good representativeness.

[0111] S3.4: Calculating the translation of a single point for:

[0112] (10)

[0113] The single-point translation amount Used to characterize the current calibrated equipment due to ambient temperature. The coefficient deviation relative to the average level caused by its own minor hardware deviations;

[0114] S3.5: Calculate the single-point offset Correction table with reference mean Simply add them together to generate the value of the device to be calibrated at the current temperature. Complete FVC correction table below for:

[0115] (11).

[0116] To verify the effectiveness of the offset method, a comparative experiment was designed, selecting 10 devices. At different temperatures, FVC calibration tables were generated using both the offset method and the traditional full flow point scanning method, and error data analysis was performed. Figure 8 This is a graph showing the error characteristics of the offset method. Figure 8 (a) is an error curve graph, showing the error between the FVC calibration table obtained by the offset method and the traditional full flow point scanning method for 10 devices. Figure 8 (b) is the error distribution histogram, which shows the comparative error analysis results of the offset method and the traditional full-flow-point scanning method in respiratory flow measurement. Statistically, the average error between the two methods is only -0.007%, and the near-zero mean indicates that the offset method is highly consistent with the traditional full-flow-point scanning method in terms of overall measurement accuracy, with no systematic bias. This finding verifies the feasibility of the offset method as a simplified calibration method—it significantly reduces the calibration workload (simplifying from full-flow-point scanning to finite-point calibration) while maintaining measurement accuracy comparable to the traditional method.

[0117] Standard deviation The excellent 0.122% accuracy further confirms the reliability of the offset method. Such small dispersion indicates that the difference between the offset method and the traditional method is very stable and exhibits minimal fluctuations across different flow points and multiple measurements. Particularly noteworthy is that 100% of the error data are within ±1%, 97.2% within ±0.5%, and 95.6% within ±0.2%, these statistical results support the effectiveness of the offset method. In engineering practice, this means that after calibrating equipment using the offset method, the measurement results are minimally different from those calibrated using the traditional full-flow-point scanning method, fully meeting the accuracy requirements of clinical respiratory monitoring.

[0118] Analysis of the normal distribution characteristics of the error data shows that the kernel density estimation curve closely matches the theoretical normal distribution, indicating that the difference between the two methods is random and unbiased, rather than a systematic bias. This distributional characteristic demonstrates that the offset method not only performs well on average across the entire flow range but also exhibits consistent performance at each flow point. The maximum value of 0.364% and the minimum value of -0.794% are both within acceptable ranges and are symmetrically distributed without significant skewness or outliers, further enhancing the reliability of the offset method. While the traditional full-flow-point scanning method offers high accuracy, it is time-consuming and labor-intensive, requiring specialized calibration equipment and individual calibration at each flow point. In contrast, the offset method only requires the relationship between the mean correction table and a single actual measured flow point to obtain results, significantly simplifying the calibration process. Analysis proves that this simplification does not come at the expense of accuracy—the difference between the two methods is far below the clinically acceptable error limit (typically ±2-3%), and even surpasses the actual needs of many clinical applications.

[0119] The error distribution histogram shows that the offset method significantly improves calibration efficiency and reduces operational complexity while maintaining comparable accuracy to the traditional full-flow-point scanning method. This validation result provides solid data support for the widespread application of the offset method in clinical settings. Simplifying the calibration process while ensuring measurement accuracy is crucial for improving equipment utilization efficiency and reducing maintenance costs.

[0120] S4: Input the FVC features into the pre-trained SVR model and output the complete PEF correction table; the specific steps are as follows:

[0121] S4.1: Using the flow approximation criterion and rounding matching, m sets of flow point pairs with clear physical correspondence were established;

[0122] In this embodiment, since direct translation is not possible, a complex function from the known state to the PEF correction coefficient must be learned through a data-driven approach. First, a bridge needs to be established between the FVC and PEF flow characteristics. Considering that the flow points defined by the two are not entirely the same, the above method is used to establish the relationship. In this embodiment, the value of m is 11, and 11 pairs of flow point pairs with clear physical correspondences are established, as shown in Table 2.

[0123] Table 2. Mapping Relationship between FVC and PEF Flow Points

[0124]

[0125] S4.2: Rapidly generate FVC features at the above m flow points using the offset method: correction coefficient vector ;

[0126] In this embodiment, these coefficients stably characterize the static gain characteristics of the device at the corresponding flow rate, forming the basis for predicting its dynamic response.

[0127] S4.3: Convert the correction coefficient vector Input the pre-trained SVR model and output the correction coefficient vector at m preset PEF flow points. Obtain the complete PEF calibration table.

[0128] In this embodiment, given the complexity of the PEF correction relationship, this application adopts a point-independent modeling strategy: a separate regression model is trained for each of the 11 PEF target points. :

[0129] (12)

[0130] This strategy allows each model to learn a unique dynamic correction relationship for its corresponding traffic point, avoiding the problems of dimensionality curse and mutual interference between different traffic point characteristics that may exist in a single complex model.

[0131] The relationship between the two is profoundly affected by the nonlinear characteristics of the equipment. In order to provide the model with richer information that reflects the underlying physical mechanism, this application has carried out in-depth feature engineering, introducing polynomial terms of the original features. To capture possible nonlinear gain, based on the principle of ultrasonic flow measurement ( ), constructing logarithmic features Sum of square root characteristics ( ), attempting to match the relationship between signal amplitude and flow rate, and adjusting the FVC coefficient Its corresponding flow point value Multiply to form an interaction term. This allows the model to perceive response differences across different flow velocity ranges. Through the above transformations, the original 11-dimensional features are expanded into a higher-dimensional, more information-rich feature vector. This laid the foundation for learning more complex models later.

[0132] In this embodiment, an optimal regressor is determined for each target point. This application presents a systematic comparative experiment on three representative machine learning regression algorithms, primarily comparing Elastic Network Regression, Support Vector Regression (SVR), and Gaussian Process Regression. Each of these three algorithms has its advantages, and Elastic Network Regression, combined with... and Regularization is suitable for feature selection and mitigating collinearity. Support Vector Regression (SVR) is based on the principle of minimizing structured risk and handles nonlinearity through kernel tricks, exhibiting strong generalization ability with small samples. Gaussian Regression is a probabilistic model that provides estimates of prediction uncertainty. Through cross-validation and performance comparison, SVR was ultimately determined to be the optimal algorithm.

[0133] Support Vector Regression (SVR) is a machine learning algorithm based on the principle of structural risk minimization, which introduces... Insensitive loss functions and kernel tricks can effectively handle small sample sizes and nonlinear regression problems. The SVR model for the i-th preset PEF flow point takes the form of:

[0134] (13)

[0135] in, The number of training samples. and It is a Lagrange multiplier and satisfies the constraints. , For penalty parameters, For bias terms, For radial basis function kernels, kernel function The radial basis function (RBF) kernel is used, specifically in the form of:

[0136] (14)

[0137] in, The kernel function parameters control the degree of influence of a single training sample.

[0138] In this embodiment, the optimization objective of SVR is to find the optimal regression hyperplane in the high-dimensional feature space, such that the deviation of all sample points from the hyperplane does not exceed a preset fault tolerance parameter. Simultaneously, minimize model complexity. This optimization problem can be formalized as the following convex optimization problem:

[0139] (15)

[0140] in These are slack variables used to handle variables that exceed... The sample points shown demonstrate the robustness of SVR to noise and outliers.

[0141] In the MATLAB implementation of this application, the SVR model is implemented through... The function is trained with the following key parameters set: kernel function type is Gaussian kernel (RBF), feature standardization is enabled, kernel scale parameter is automatically optimized, and box constraint parameter is set to 1. These settings ensure that the model can maintain good generalization performance while ensuring prediction accuracy.

[0142] To ensure that the prediction results conform to both statistical patterns and engineering physical constraints, this application designs a hierarchical optimization strategy, including two levels: local optimization and global optimization. The local optimization layer mainly refines the prediction results for individual flow points, and for points with relatively weak prediction performance (judged by...),... The algorithm employs a flow continuity constraint, utilizing the predicted values ​​of adjacent flow points and performing weighted smoothing based on flow distance.

[0143] (16)

[0144] in, and These represent the distances between the current flow point and its immediate neighbors. The final predicted value is obtained by a weighted fusion of the original predicted value and the constrained predicted value.

[0145] (17)

[0146] Simultaneously, an outlier detection method based on the interquartile range (IQR) is employed to identify and correct statistical outliers in the predicted sequence, replacing them with median estimates to enhance the model's robustness. The global optimization layer targets traffic points that consistently perform poorly (…). ), utilizing cross-point information for reinforcement learning, specifically by starting from all high-performance points ( Extract features and label data from the dataset and train a global gradient boosting model.

[0147] (18)

[0148] in For weak learners (decision trees) These are the weighting coefficients. This represents the number of weak learners. The predictions from the global model are then merged with the original predictions in a 4:6 ratio.

[0149] (19)

[0150] Experimental results show that Support Vector Regression performs well in the PEF correction prediction task. The main reason is that the data size of this study (155 valid samples) is exactly within the advantageous range of the SVR algorithm, and the RBF kernel function can effectively capture the high-dimensional nonlinear relationship after feature engineering. The principle of minimizing structured risk effectively controls the model complexity and reduces the risk of overfitting. SVR has strong robustness to scale changes and noise interference of input data and is suitable for practical engineering applications.

[0151] To systematically evaluate the performance of different machine learning algorithms in the PEF correction table prediction task, this application designed a complete comparative experimental framework. The experiment was implemented using the MATLAB platform. By reading the preprocessed device data file, 11 sets of flow point correspondences were constructed for disposable breathing tubing configurations. In the data preprocessing stage, the FVC flow points and PEF flow points were aligned using an exact matching algorithm, resulting in a standardized dataset containing 155 valid samples.

[0152] First, by constructing a scatter regression plot between predicted and actual values, the mapping ability of each model on the full dataset is visually demonstrated, such as... Figure 9 The figure shown is a comparison of the regression relationship between predicted values ​​and actual values ​​from different models. Figure 9 (a) is a scatter regression plot of elastic network regression. As can be seen from the plot, the scatter points are relatively scattered, especially in the interval where the nonlinear characteristics are obvious, there is a large prediction bias. This indicates that the linear hybrid strategy has limitations in dealing with such highly nonlinear data problems that are affected by multiple factors (circuit components, pipeline structure and installation method). Figure 9 (b) is a scatter plot of support vector regression, where the scatter points are most closely clustered around... On both sides of the ideal diagonal, due to the use of radial basis function kernels, it can capture the smallest nonlinear changes in the original FVC signal, thus achieving the most accurate global reconstruction under the condition of low feature correlation; Figure 9 (c) is a scatter plot of the Gaussian process regression. Although the overall trend remains consistent, there are some fluctuations in certain flow margins, reflecting the sensitivity of the kernel function to factors influencing the data. The regression relationship between the predicted and actual values ​​from the three models, along with the corresponding coefficients of determination, is also shown. The comparison shows that support vector regression is the best performing model.

[0153] To further quantitatively evaluate the compensation effect of the model, this application conducted a detailed statistical analysis of the prediction residuals of elastic network regression, support vector regression (SVR), and Gaussian process regression. The mean absolute error (MAE) and residual standard deviation (Std) of each model were extracted in the full range of experiments, and the comparison results are shown in Table 3.

[0154] Table 3 Comparison of statistical indicators of prediction residuals from different models

[0155]

[0156] Experimental data show that the MAE of the SVR model is only 0.0025, which is about 81.3% higher than that of the linearized elastic network model (0.0134). This indicates that SVR, through the eigenmapping of higher-order kernel functions, can more accurately fit the physical response of ultrasonic flow meters under various nonlinear operating conditions. Its residual standard deviation (Std=0.0043) is also the lowest among the three, proving that SVR has extremely high predictive consistency when dealing with flow fluctuations.

[0157] S5: Burn the FVC calibration table and PEF calibration table to the device, verify the results, and the calibration is complete.

[0158] S6: Experimental verification:

[0159] The core objective of the experiment was to quantitatively evaluate the optimized scheme from two dimensions: efficiency and accuracy. First, in terms of efficiency, the experiment precisely measured the total operation time required for the new scheme to complete the full calibration of a single device (FVC and PEF indicators, one-time piping) and directly compared it with the traditional method. As shown in Table 4, each calibration step was optimized to a certain extent, and the efficiency was improved by 85%-90%. The new scheme reduced the calibration time from hours in the traditional scheme to minutes.

[0160] Table 4 Comparison of Time Consumption for New and Old Calibration Methods

[0161]

[0162] In terms of accuracy, after using the new scheme, the measured values ​​of the viewing device at multiple flow points were calculated, and the relative error between them and the standard flow meter readings was calculated. The test results of PEF are shown in Table 5. As can be seen from Table 5, the PEF index and the average error of each flow point for each device are within an extremely low error range, and the FVC index is also within an extremely low error range. Due to the large amount of data, it will not be elaborated here. Furthermore, the devices that have passed the calibration process of the new scheme have passed the registration test conducted by the Chongqing Medical Device Quality Inspection Center in accordance with the YY / T 1438-2016 standard.

[0163] Table 5 Summary of PEF Index Measurement Errors

[0164]

[0165] In summary, the integrated optimization scheme of "installation fixture + offset method + SVR prediction model" proposed in this application fundamentally reconstructs the calibration process by combining hardware benchmark unification with software intelligent prediction, thereby improving the efficiency and accuracy of calibration.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A rapid calibration system for an ultrasound pulmonary function testing device, characterized in that, Includes a mounting fixture for positioning and installing the device to be calibrated (8), the mounting fixture including a base (1); A linear cylinder (2) and a passive positioning plug (3) are installed on the base (1). The telescopic end of the linear cylinder (2) is connected to one end of the elastic push-pull module, and the other end of the elastic push-pull module is connected to the active positioning plug (4). When the calibration device (8) is installed, the components of the calibration device (8) are arranged between the passive positioning plug (3) and the active positioning plug (4).

2. The rapid calibration system for an ultrasound pulmonary function testing device according to claim 1, characterized in that, The elastic push-pull module includes a limiting block (5) installed on the base (1), and the limiting block (5) has a through hole arranged along the movement direction of the linear cylinder (2); A sleeve (6) is slidably disposed in the through hole, and a spring (7) is installed in the sleeve (6). One end of the sleeve (6) is connected to the active positioning plug (4), one end of the spring (7) abuts against one end of the active positioning plug (4), and the other end of the spring (7) is connected to the front end of the floating joint rod (9). The front end of the floating joint rod (9) can slide along the sleeve (6), and the rear end of the floating joint rod (9) is connected to the telescopic end of the linear cylinder (2).

3. The rapid calibration system for an ultrasound pulmonary function testing device according to claim 1, characterized in that, The device to be calibrated (8) includes a pipe housing (801) and an ultrasonic transducer (802), wherein the ultrasonic transducer (802) is pressed into the pipe housing (801) by means of a mounting fixture.

4. The rapid calibration system for an ultrasound pulmonary function testing device according to claim 1, characterized in that, It also includes standard flow generators and processors; During equipment calibration, the standard flow generator is connected to the device to be calibrated (8) and provides a stable airflow to the device to be calibrated (8). The processor is used to calibrate the device based on the data collected by the flow sensor of the device to be calibrated (8).

5. A rapid calibration method for an ultrasound pulmonary function testing device, characterized in that, The rapid calibration system for the ultrasound pulmonary function testing device according to any one of claims 1-4 comprises the following steps: S1: Install the equipment to be calibrated using the mounting fixture and verify that the reference distance between the two pressure points is acceptable; S2: Connect to a standard flow generator to collect single-point data at characteristic flow points; S3: Using the offset method, a complete FVC correction table at any temperature is synthesized by combining the reference mean FVC correction table and the collected single-point offset. S4: Input the FVC features into the pre-trained SVR model and output the complete PEF correction table; S5: Burn the FVC calibration table and PEF calibration table to the device, verify the results, and the calibration is complete.

6. The rapid calibration method for an ultrasound pulmonary function testing device according to claim 5, characterized in that, The specific steps in step S1 are as follows: S1.1: Place the pipe housing (801) and the ultrasonic transducer (802) between the two passive positioning plugs (3) and the active positioning plug (4) to complete the initial positioning and acoustic axis alignment of the ultrasonic transducer (802); S1.2: Start the linear cylinder (2), the linear cylinder (2) pushes the elastic push-pull module and the ultrasonic transducer (802) to move forward along the guide axis, and the ultrasonic transducer (802) enters the pipe shell (801). S1.3: The linear cylinder (2) continues to advance, the spring (7) is compressed, and a linearly increasing pressing force is generated, which smoothly and accurately presses the ultrasonic transducer (802) into the designed position of the pipe shell (801); S1.4: After the press-fit is in place, the linear cylinder (2) briefly maintains pressure in the air circuit to ensure that the ultrasonic transducer (802) is stably in place and completes the first installation of the equipment (8) to be calibrated.

7. The rapid calibration method for an ultrasonic pulmonary function testing device according to claim 5, characterized in that, The specific steps in step S3 are as follows: S3.1: Under production conditions with the application of mounting fixtures, collect FVC calibration tables for multiple devices at various ambient temperatures; S3.2: Calculate the arithmetic mean of multiple FVC correction tables at each flow point to generate a reference mean FVC correction table decoupled from a specific temperature. ; S3.3: Place the new device to be calibrated at the current ambient temperature. Next, select a specific flow point. Perform a ventilation test and compare the results to determine the operating conditions of the device to be calibrated at the current temperature. Below, at the flow point Measured correction coefficient at the location ; S3.4: Calculate the translation of a single point for: The single-point translation amount Used to characterize the current calibrated equipment due to ambient temperature. The coefficient deviation relative to the average level caused by its own minor hardware deviations; S3.5: Calculate the single-point offset Correction table with reference mean Simply add them together to generate the value of the device to be calibrated at the current temperature. Complete FVC correction table below for: .

8. The rapid calibration method for an ultrasound pulmonary function testing device according to claim 5, characterized in that, The specific steps in step S4 are as follows: S4.1: Using the flow approximation criterion and rounding matching, m sets of flow point pairs with clear physical correspondence were established; S4.2: Rapidly generate FVC features at the above m flow points using the offset method: correction coefficient vector ; S4.3: Convert the correction coefficient vector Input the pre-trained SVR model and output the correction coefficient vector at m preset PEF flow points. Obtain the complete PEF calibration table.

9. A rapid calibration method for an ultrasonic pulmonary function testing device according to claim 8, characterized in that, In step S4, m independent SVR models are trained for each of the m preset PEF flow points; the SVR model for the i-th preset PEF flow point is in the following form: ； in, The number of training samples. and It is a Lagrange multiplier and satisfies the constraints. , For penalty parameters, For bias terms, For radial basis function kernels, kernel function The radial basis function (RBF) kernel is used, specifically in the form of: ； in, The kernel function parameters control the degree of influence of a single training sample.

Images