Helicobacter pylori breath test signal denoising fitting method and system

Abstract

The application discloses a Helicobacter pylori breath test signal denoising fitting method and system, and belongs to the technical field of medical detection and signal processing. The method processes the continuous monitoring signal of the 13CO2 / 12CO2 isotope ratio in exhalation: firstly, the original signal is subjected to band-pass digital filtering to suppress high-frequency noise; then, low-frequency drift is eliminated through adaptive baseline correction to generate a drift-removed signal; subsequently, a robust peak value recognition algorithm is used to extract physiological response characteristics; finally, the identified peak value is taken as the center, a nonlinear function with physiological constraints is used for local fitting, and the Delta Over Baseline value is calculated based on the generated fitting curve with high precision. The application effectively solves the signal distortion problem caused by breathing fluctuations, instrument noise and the like in in vivo detection, realizes high-fidelity restoration of the Helicobacter pylori metabolic response, and significantly improves the detection accuracy and reliability of the breath test.

Description

Technical Field

[0001] This invention relates to the field of medical testing and signal processing technology, specifically to a method and system for noise reduction and fitting of Helicobacter pylori breath test signals. Background Technology

[0002] Helicobacter pylori (Hp) is a significant pathogenic factor in chronic gastritis, peptic ulcers, and even gastric cancer. Accurate identification of its infection status is crucial for clinical diagnosis and treatment. Among various detection methods, the urea breath test (UBT) has become the preferred non-invasive method in current clinical practice due to its non-invasiveness, high sensitivity, and high specificity. This technology is based on the urease activity unique to Hp, and involves the subject orally ingesting isotope-labeled urea (such as ¹³C or ¹³C). 4 C) If Hp infection is present, urea is rapidly hydrolyzed to generate isotopically labeled carbon dioxide (CO2), which is then exhaled through the lungs. The concentration increment is detected by high-precision mass spectrometry or spectroscopy equipment, and the Delta OverBaseline (DOB) value is calculated based on this to determine the infection status.

[0003] However, with the increasing demands for detection accuracy and reproducibility in clinical practice, UBT has revealed several deep-seated technical bottlenecks in practical applications. While existing mainstream methods have been optimized in terms of sampling strategies or detection principles, they generally neglect the complex interference issues faced by the raw expiratory signal during acquisition and transmission. For example, patent CN117517238A proposes to dynamically track the DOB value trend by sampling multiple times at three time points (10, 20, and 30 minutes) after medication to avoid the risk of false negatives at fixed sampling times due to individual metabolic differences. This method does improve the individual adaptability of the test to some extent, but it still relies on the comparison of samples at discrete time points and does not address the noise suppression mechanism of the raw signal itself in a single measurement. In other words, even if the sampling time is more reasonable, if the signal obtained at each time point contains significant random fluctuations, baseline drift, or respiratory rhythm interference, the DOB calculation will still be inaccurate. Furthermore, multi-point sampling further increases the operational burden and reagent consumption, making it difficult to meet the needs of efficient and low-cost clinical practice.

[0004] The invention patent with announcement number CN110487731B shifts to a high-precision spectral detection approach. It utilizes laser ablation to generate CN radicals and analyzes isotopic abundance using a multi-channel spectrometer. While this theoretically improves detection resolution, its signal processing focuses on frequency and intensity analysis of static spectral features, lacking the ability to systematically model and filter out the dynamic characteristics inherent in exhalation signals as continuous time series, such as low-frequency baseline drift, high-frequency electronic noise, and non-steady-state fluctuations caused by irregular breathing. This approach does not incorporate adaptive filtering or nonlinear fitting mechanisms, making it difficult to effectively separate the true physiological response signal from instrument background noise or environmental disturbances. This is particularly problematic in the elderly, children, or those with limited respiratory function, where the signal-to-noise ratio significantly decreases, easily leading to misjudgments.

[0005] Existing UBT systems, while pursuing detection sensitivity and ease of operation, have failed to simultaneously build matching signal cleansing and feature reconstruction capabilities. Traditional methods often treat signals as idealized discrete data points, ignoring the fact that exhalation is essentially a continuous dynamic process influenced by multiple coupled physiological and engineering factors. In this process, noise not only manifests as random perturbations superimposed on the signal, but also often exists as slowly drifting baseline shifts or spurious peaks overlapping with physiological rhythms. If such interference is not specifically addressed, it will directly distort the identification of peak positions and amplitudes, leading to deviations in DOB value calculation. Consequently, simply increasing the number of sampling points or improving hardware resolution cannot fundamentally solve the problem of signal quality degradation; on the contrary, data redundancy may exacerbate the uncertainty of subsequent analysis.

[0006] How to construct an integrated signal processing framework that combines digital filtering, adaptive baseline correction, robust peak identification, and nonlinear curve fitting to effectively suppress multi-source interference while preserving real physiological characteristics, thereby significantly improving the accuracy of DOB value calculation and the clinical reliability of detection results, has become a key challenge and an urgent technical problem for those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for noise reduction and fitting of Helicobacter pylori breath test signals; it solves the technical problem that the calculation of critical diagnostic value (DOB) is inaccurate in existing Helicobacter pylori breath tests due to interference from multiple sources of noise in the original signal.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A method for noise reduction and fitting of Helicobacter pylori breath test signals includes the following steps:

[0010] The time-series data of the ¹³CO₂ / ¹²CO₂ isotope ratio in the exhaled air of the subjects after drug administration were continuously collected using an infrared spectroscopy sensor at a sampling frequency of no less than 10 Hz, forming the raw signal. ,in This represents a time variable starting from the moment the medication was taken;

[0011] For the original signal Apply bandpass digital filtering to obtain a preliminary purified signal. ;

[0012] Based on the preliminary purification signal Constructing an adaptive baseline model from low-frequency components And perform baseline correction operation to generate drift-free signal. Subsequently, in the de-drift signal A robust peak identification algorithm based on morphological constraints is executed to determine the peak time point of the physiological response. and their corresponding amplitudes ;

[0013] At the peak time point Centered on, within a preset time window Internally, a nonlinear function with physiological prior constraints is used to de-drift signals. Perform local fitting to generate a fitted curve. The DOB value is calculated based on the fitted curve.

[0014] Furthermore, the infrared spectral sensor is a dual-channel non-dispersive infrared (NDIR) detector, with the center wavelength of its first channel set at 4.26 μm, used to detect the ¹²CO₂ absorption intensity. The center wavelength of the second channel was set to 4.35 μm to detect the ¹³CO₂ absorption intensity. The original signal Defined as isotope ratio And convert it to the δ¹³C value using the following formula:

[0015] ;

[0016] in, To measure the isotope ratio, The ¹³C / ¹²C ratio of PDB (Pee Dee Belemnite) carbonate, according to international standard, is 0.0112372; the original signal That is Time series.

[0017] In a preferred embodiment of the present invention, the bandpass digital filtering process employs a fifth-order Butterworth bandpass filter with a low cutoff frequency. Set to 0.005 Hz, high cutoff frequency. Set to 0.5 Hz; the transfer function of this filter. Defined by the following formula:

[0018]

[0019] Among them, coefficient and The discretization is based on the standard Butterworth polynomial discretized using the bilinear transformation method. For Laplace variables, the sampling period is... s; The filtering operation modifies the original signal Zero-phase forward-backward filtering is used to eliminate phase distortion.

[0020] Furthermore, the adaptive baseline model The construction process is as follows:

[0021] For the initial purification signal Performing Empirical Mode Decomposition (EMD) yields several Intrinsic Mode Functions (IMFs) and a residual term. ;

[0022] Calculate the center frequency of each IMF component ( ),in The instantaneous frequency is obtained by Hilbert transform and then averaged over time.

[0023] Set baseline determination threshold Hz, satisfying all IMF components and residuals Add them together to form the initial baseline estimate. Finally, regarding Apply sliding window mid-range filtering, window length s, to obtain the final adaptive baseline model .

[0024] As one of the key innovations of this invention, the robust peak identification algorithm includes the following sub-steps:

[0025] In the drift signal Calculate the first derivative With the second derivative Set amplitude threshold ‰, only those that meet the requirements are retained. The time interval; search within this interval for... and The local maximum point; introducing respiratory rhythm constraints: if the time interval between two adjacent candidate peaks is... If the amplitude is less than 8 seconds or greater than 120 seconds, those with smaller amplitudes are discarded; the peak value with the largest amplitude among the remaining candidate peak values ​​is selected as the physiological response peak value, and its corresponding time is [missing value]. The amplitude is .

[0026] Furthermore, the nonlinear function is a modified gamma distribution function, and its mathematical expression is:

[0027] ;

[0028] in, This is the amplitude scaling factor. This is the initial delay time. It is a time constant. For shape parameters, The residual offset; within the preset time window s, The fitting process employs the Levenberg-Marquardt nonlinear least squares algorithm, with the objective function being:

[0029] ;

[0030] And the following physiological prior constraints are imposed: s, s, , ‰, .

[0031] As another key feature of the present invention, the calculation of the DOB value is based on a fitted curve. Instead of the original discrete sampling points, the specific formula is:

[0032] ;

[0033] in, The baseline duration is 300 s; the integration operation uses the composite Simpson's rule, with a step size of... This calculation method effectively avoids baseline mean fluctuations caused by single-point noise, while using the smoothness of the fitted curve to suppress random disturbances at the peak.

[0034] In addition, the present invention also provides a Helicobacter pylori breath test signal noise reduction and fitting system, the system including a signal acquisition module, a digital filtering unit, a baseline correction unit, a peak recognition unit, a nonlinear fitting unit and a DOB calculation unit; each unit is interconnected through an internal bus and the above method flow is uniformly scheduled and executed by a central processing unit.

[0035] The signal acquisition module includes a dual-channel NDIR sensor, a temperature compensation circuit, a pressure sensor, and an analog-to-digital converter. The NDIR sensor has an optical cavity length of 15 cm, a pulse-modulated infrared LED light source with center wavelengths of 4.26 μm and 4.35 μm, and a full width at half maximum (FWHM) of 50 nm. The analog-to-digital converter is a 24-bit Σ-Δ ADC with a sampling rate of 100 SPS, which is downsampled to 10 Hz by software. The temperature compensation circuit monitors the cavity temperature in real time based on a platinum resistance thermometer (PT1000) and corrects the CO2 absorption coefficient for temperature based on the NIST standard gas database.

[0036] The digital filtering unit is implemented in an embedded microcontroller with an ARM Cortex-M7 architecture, a main frequency of 480 MHz, and a built-in hardware floating-point unit (FPU). The filter coefficients are pre-stored in read-only memory (ROM), and the filtering operation adopts a fixed-point optimization algorithm to reduce memory usage. Zero-phase filtering is achieved by calling the same filter twice to process the forward and reverse sequences respectively.

[0037] When the baseline correction unit performs EMD decomposition, it adopts an improved screening stopping criterion: it terminates when the standard deviation of the IMF obtained from three consecutive screenings is less than 0.01, or when the number of screenings reaches 10. The Hilbert transform is implemented by combining the Fast Fourier Transform (FFT) with the analytic signal construction method, and the number of FFT points is... The sliding window value filter uses a circular buffer structure, with a constant memory footprint. byte.

[0038] The peak identification unit employs a five-point central difference scheme in the calculation of the first and second derivatives, and its discrete expression is as follows:

[0039]

[0040]

[0041] in, Indicates the first The time for each sampling point s represents the sampling interval; the time interval thresholds of 8 s and 120 s in the respiratory rhythm constraint are based on large-scale clinical data statistics including 600 Hp-positive patients and 600 Hp-negative healthy subjects, covering the expiratory metabolic kinetic range of 95% of the population.

[0042] The nonlinear fitting unit adopts the following strategy when initializing its parameters: , s, , Start delay time initial value By peak time Multiply by an empirical coefficient Obtain, that is ,in The value range is from 0.2 to 0.4 to ensure... fall into Within the physiological prior constraints of s; the damping factor of the Levenberg-Marquardt algorithm. The initial value is set to 0.01, and is dynamically adjusted based on the change in the sum of squared residuals in each iteration; the maximum number of iterations is set to 100, with a convergence tolerance of [missing value]. .

[0043] When performing integration operations, the DOB calculation unit calculates the fitted curve. exist Sampling is performed at 1-second intervals within the interval, resulting in a total of 301 points. The compound Simpson's formula is then applied.

[0044] ;

[0045] in This numerical integration method avoids direct integration of the original noise signal while ensuring accuracy.

[0046] Furthermore, the system of the present invention also includes a quality control module for real-time evaluation of the effectiveness of each stage of signal processing; the quality control module performs the following checks:

[0047] (1) Original signal During the baseline period Standard deviation within s It should be less than 1.5‰; otherwise, it indicates unstable breathing or equipment malfunction.

[0048] (2) Preliminary purification signal energy concentration It should be greater than 0.6; otherwise, the signal is considered to have no significant physiological response.

[0049] (3) Root mean square error of the fitting residuals It should be less than 0.8‰, otherwise it will trigger a refit or manual review process.

[0050] In a preferred embodiment of the present invention, the system is integrated into a portable UBT detector, with overall dimensions of 25cm × 18cm × 10cm and a weight not exceeding 2.5kg; the power supply uses a rechargeable lithium-ion battery with a capacity of 10,000mAh, supporting continuous operation for more than 8 hours; the human-machine interface is a 7-inch capacitive touchscreen, running a Linux-based real-time operating system; data storage uses eMMC flash memory with a capacity of 32GB, supporting export in CSV and DICOM formats.

[0051] When the number of candidate physiological response peaks is identified When the value is greater than or equal to 2, the bimodal competitive fitting mechanism is activated; the bimodal competitive fitting mechanism is executed according to the following steps: B1. Initial determination of the existence of bimodal peaks: Select the peak with the largest amplitude from all candidate peaks. The second largest peak value Form peak pairs to be evaluated; calculate peak pair time intervals. ;like Falling within the preset effective bimodal interval If the signal is within the range of B1, it is determined to be a potential bimodal signal, and the process proceeds to step B2; otherwise, it is determined to be a unimodal signal, and the unimodal fitting process continues. It lasts for 60 seconds. The duration is 1800 seconds; B2. Construct a bimodal metabolic kinetic fusion model: establish a bimodal competitive fitting function. This function is a weighted sum of two modified gamma distribution functions with independent metabolic parameters, and its expression is:

[0052] ;

[0053] in, and These correspond to the peak values ​​respectively. and The modified gamma distribution function, in the form of claim 7 same; and These are the parameter vectors for the two sub-models; and The peak weighting coefficient satisfies and The initial value is determined by the ratio of the amplitudes of the two peaks:

[0054]

[0055] B3. Collaborative Fitting and Model Competition: In Extended Time Windows Internally, for the drift signal Simultaneously perform single-peak model (by (centered) and bimodal model The fit; where, Second, Seconds; the fitting process uses the constrained Levenberg-Marquardt algorithm and introduces inter-peak coupling constraints: Second, Seconds; B4. Model selection based on information criterion: Calculate the Bayesian information criterion values ​​for unimodal and bimodal fitting results respectively. and :

[0056] ;

[0057] in, Number of model parameters (single-peak model) Bimodal model ), This represents the total number of data points within the fitting window. Let S be the sum of squared residuals of the model fit; if If so, the bimodal model is determined to be superior, and it is selected. The final fitted curve is used; otherwise, the unimodal model is considered superior. This is a tolerance factor, with a value of 2.0, used to prioritize simpler models unless the bimodal model has a sufficiently significant fit advantage; B5. DOB calculation in bimodal mode: When selecting a bimodal model At that time, the final DOB value is calculated as follows:

[0058] ;

[0059] in, This represents the global maximum value of the bimodal fitted curve within the time interval.

[0060] Furthermore, in step B1, after identifying a potential bimodal signal and before executing step B2, a peak purity verification step is added: calculating the peak values ​​respectively. and Local signal-to-noise ratio in the vicinity If any peak value If the bimodal competitive fitting fails, the process reverts to unimodal fitting; the local signal-to-noise ratio... Defined as peak amplitude With peak time A 60-second interval centered on [the specified location] Inside, exclude De-drift signal after the second range Standard deviation The ratio, i.e. The effective bimodal interval The determination was based on data on Helicobacter pylori metabolic kinetics and gastric emptying physiology, among which Seconds are used to eliminate breathing artifacts. The second covers the subsequent metabolic peak that may be caused by delayed gastric emptying.

[0061] The nonlinear fitting unit is configured to: when the peak identification unit identifies the number of candidate physiological response peaks When the value is greater than or equal to 2, the bimodal competitive fitting method as described in claim 14 or 15 is executed.

[0062] This invention constructs an end-to-end signal reconstruction closed loop constrained by physiological mechanisms. This closed loop isolates the effective frequency band through bandpass filtering, eliminates slow-varying drift through adaptive baseline correction, identifies real physiological events through morphological peak recognition, and finally achieves smoothing and noise reduction of feature extraction through nonlinear fitting with prior constraints. These four interconnected steps ensure that the signal features upon which the DOB value calculation depends are faithful to the actual physiological process while minimizing interference from engineering and physiological noise.

[0063] This invention solves the technical problem of inaccurate DOB values ​​caused by multi-source interference in the original signal of the Helicobacter pylori breath test. It possesses comprehensive advantages such as high accuracy, strong robustness, low cost, and easy deployment, making it suitable for medical institutions at all levels. It is particularly beneficial for promoting high-precision Hp screening in grassroots areas and resource-limited regions, demonstrating significant clinical value and industrialization prospects.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] This invention overcomes the limitations of traditional methods that treat exhalation signals as discrete sampling points, reconstructing them as a continuous dynamic process constrained by physiological mechanisms. Through four interconnected steps—bandpass filtering, adaptive baseline correction, robust peak identification, and nonlinear fitting—the system can extract the true physiological response curve characterizing Helicobacter pylori metabolic activity from the original signal contaminated by complex noise with high fidelity. Based on this smooth, noise-resistant fitted curve, the DOB value is calculated, fundamentally suppressing peak misjudgment and baseline fluctuations caused by random noise, baseline drift, and respiratory artifacts. This results in diagnostic results that are more faithful to the true infection state, significantly reducing the misjudgment rate.

[0066] The adaptive baseline correction technology employed in this invention can dynamically track and eliminate slow-varying drift in the signal, while the peak recognition algorithm, which integrates morphological and physiological rhythm constraints, can effectively identify and eliminate transient spurious peaks caused by coughing, irregular breathing, etc. Particularly for cases with complex metabolic dynamics, the introduced bimodal competitive fitting mechanism can intelligently identify and accurately fit bimodal metabolic signals through a data-driven approach, avoiding the systematic bias of traditional single-peak models in such scenarios. These characteristics enable the system to maintain stable and reliable detection performance even when facing subjects with inherently low signal-to-noise ratios, such as the elderly, children, or those with respiratory insufficiency, thus expanding the reliable application range of the urea breath test.

[0067] This invention improves performance while also ensuring ease of engineering implementation and cost-effectiveness. The entire signal processing flow can be implemented using a general-purpose embedded processor, eliminating the need for expensive dedicated computing hardware. For the sensor, a cost-effective dual-channel non-dispersive infrared detector is preferred. High-precision measurement results are achieved through advanced signal processing technology, avoiding the use of high-end mass spectrometry or laser spectrometers. The system can be highly integrated into portable devices, with a simple operation process and complete real-time quality control functions. This makes it particularly suitable for deployment and promotion in primary healthcare institutions and large-scale screening scenarios, providing a solid technical foundation for achieving universal high-precision diagnosis of Helicobacter pylori infection. Attached Figure Description

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0069] Figure 1 This is an overall architecture diagram of the system described in this invention.

[0070] Figure 2 This is an overall flowchart of the method described in this invention.

[0071] Figure 3 This is a simplified flowchart of the method described in this invention.

[0072] Figure 4 This is one of the system operation interfaces described in this invention.

[0073] Figure 5 This is the second system operation interface described in this invention.

[0074] Figure 6 This is the third type of system operation interface described in this invention.

[0075] Figure 7 This is the fourth type of system operation interface described in this invention.

[0076] Figure 8 This is the fifth system operation interface described in this invention.

[0077] Figure 9 This is the sixth system operation interface described in this invention. Detailed Implementation

[0078] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0079] The following is in conjunction with the appendix Figures 1-3 The embodiments of the present invention will be described in detail below.

[0080] Example 1: This example discloses a noise reduction and fitting method for Helicobacter pylori breath test signals. It constructs a four-level cascaded signal processing architecture that integrates digital filtering, adaptive baseline correction, robust peak identification, and nonlinear dynamic curve fitting to perform high-fidelity reconstruction of continuously acquired isotope-labeled carbon dioxide concentration time series.

[0081] The following will provide a complete, sufficient and reproducible engineering description of the technical solution of this invention, taking into account specific hardware configurations, algorithm parameters, numerical implementation details and clinical validation data.

[0082] In this embodiment, the signal acquisition module serves as the front-end sensing unit of the entire system. It consists of a dual-channel non-dispersive infrared (NDIR) detector, with the center wavelength of its first channel set at 4.26 μm for detecting the ¹²CO₂ absorption intensity. The center wavelength of the second channel was set to 4.35 μm to detect the ¹³CO₂ absorption intensity. The NDIR sensor has an optical cavity length of 15 cm and uses a pulse-modulated infrared LED as its light source. The full width at half maximum (FWHM) of both channels is 50 nm, ensuring sufficient spectral selectivity near the absorption peak of the target gas.

[0083] The analog-to-digital converter (ADC) uses a 24-bit Σ-Δ type ADC with an original sampling rate of 100 SPS. After software downsampling, the output time series has an effective sampling frequency of 10 Hz, i.e., the sampling period. The temperature compensation circuit uses a platinum resistance thermometer (PT1000) to monitor the cavity temperature in real time and, based on the temperature dependence model of the CO2 absorption coefficient in the NIST standard gas database, performs temperature compensation. and Dynamic corrections are performed to eliminate measurement drift caused by ambient temperature fluctuations. A pressure sensor synchronously acquires the intracavitary pressure to normalize the gas density, ensuring the physical consistency of the isotope ratio calculations.

[0084] Original signal The time series defined as the δ¹³C value is calculated using the following formula:

[0085] ;

[0086] in To measure the isotope ratio, The ¹³C / ¹²C ratio for PDB (Pee Dee Belemnite) carbonate, conforming to the international standard, is 0.0112372. The sequence is the original input signal for subsequent signal processing. Its time variable Data was collected continuously for 60 minutes (3600 seconds) from the time the subject took the medication.

[0087] Digital filter unit A bandpass digital filter is applied. A fifth-order Butterworth bandpass structure with a low cutoff frequency is used. Hz, high cutoff frequency Hz corresponds to the typical physiological frequency band of Helicobacter pylori metabolic response (approximately 0.005–0.5 Hz, i.e., a period of 200 seconds to 2 seconds). The continuous-domain transfer function of the filter. for:

[0088] ;

[0089] Where the coefficient and The sampling period is obtained by discretizing the standard Butterworth polynomial using the bilinear transform method. s.

[0090] To avoid phase distortion affecting peak timing localization, the filtering operation adopts a zero-phase forward-backward filtering strategy: firstly, ... The intermediate sequence is obtained by inputting the filter in chronological order. Then Reverse input to the same filter to obtain The final output is This process is implemented in an embedded microcontroller based on the ARM Cortex-M7 architecture with a clock frequency of 480 MHz. It has a built-in hardware floating-point unit (FPU), filter coefficients are pre-frozen in ROM, and the operation uses the Q15 fixed-point format to reduce memory usage and power consumption.

[0091] The baseline correction unit performs initial purification on the signal. Perform adaptive baseline modeling.

[0092] The process first involves Empirical Mode Decomposition (EMD) is performed using an improved screening stopping criterion:

[0093] The process terminates when the standard deviation of the IMF obtained from three consecutive screenings is less than 0.01, or when the number of screenings reaches 10, in order to avoid over-decomposition.

[0094] The decomposition result contains N eigenmode functions ( and a residual term For each Constructing analytic signals ,in The Hilbert transform is represented by the Fast Fourier Transform (FFT), and the number of FFT points is... Instantaneous frequency Its time average is the center frequency. Set baseline determination threshold. Hz, satisfying all IMF components and residuals Add them together to form the initial baseline estimate. .

[0095] right Apply sliding window mid-range filtering, window length s (i.e., 1200 sampling points) is implemented using a circular buffer structure, with a constant memory footprint. Bytes (float type), output as the final adaptive baseline model De-drift signal .

[0096] Subsequently, the peak recognition unit in A robust peak identification algorithm based on morphological constraints is executed. First, the first derivative is calculated using a five-point central difference scheme. With the second derivative Its discrete expression is:

[0097] ;

[0098] ;

[0099] in s; Set amplitude threshold ‰, only retained The time interval is selected to exclude low-amplitude noise disturbances. Within this interval, the search is performed to satisfy... and The local maxima are selected as candidate peaks. Then, a respiratory rhythm constraint is introduced: if the time interval between any two adjacent candidate peaks is... s or If the amplitude is s, then those with smaller amplitudes are removed. This threshold range is based on a statistical analysis of expiratory metabolic kinetics of 1200 clinical subjects (including 600 Hp-positive and 600 Hp-negative), covering the physiological response rhythm of 95% of the population. The peak value with the largest amplitude among the remaining candidate peaks is selected as the physiological response peak, and its corresponding time is s. The amplitude is .

[0100] In determining Then, the nonlinear fitting unit uses Centered on, within a preset time window Internal Perform local fitting, where s, The fitting function uses a modified gamma distribution function:

[0101] ;

[0102] This function has a clear physiological meaning: Indicates the time delay in the onset of drug absorption and metabolism. The time constant that reflects metabolic kinetics The steepness of the rise and the gentleness of the fall of the control curve. This is the peak amplitude scaling factor. This represents the residual offset. The fitting process uses the Levenberg-Marquardt nonlinear least squares algorithm, with the objective function being:

[0103] ;

[0104] And the following physiological prior constraints are imposed: s, s, , ‰, .

[0105] The parameter initialization strategy is as follows: , s, , Start delay time initial value By peak time Multiply by empirical coefficient Obtain, that is To ensure it falls into Within the constraints of s. The damping factor of the Levenberg-Marquardt algorithm. The initial value is set to 0.01, and is dynamically adjusted based on the change in the sum of squared residuals in each iteration; the maximum number of iterations is set to 100, with a convergence tolerance of [missing value]. After fitting is complete, a smooth, noise-resistant fitting curve is generated. .

[0106] The DOB calculation unit is based on the fitted curve. Calculate the final diagnostic indicators. The DOB value is defined as:

[0107] ;

[0108] in s represents the baseline period duration. Integration is performed using the compound Simpson's rule, with a step size... s, in There are a total of 301 sampling points within the interval, and their numerical expressions are as follows:

[0109]

[0110] This method avoids direct integration of the original noise signal and significantly improves the stability of the baseline mean by utilizing the smoothness of the fitted curve.

[0111] The quality control module runs throughout the entire signal processing flow, evaluating the effectiveness of each stage in real time. It performs three key checks:

[0112] (1) Original signal During the baseline period Standard deviation within s It should be less than 1.5‰, if If the reading is ‰, it indicates that the subject's breathing is unstable, the equipment is leaking, or the sensor is malfunctioning, and the system will automatically prompt for a retest.

[0113] (2) Preliminary purification signal Energy concentration It should be greater than 0.6, if If no significant physiological response is observed, the signal is considered to be a false negative or invalid sample.

[0114] (3) Root mean square error of the fitting residuals It should be less than 0.8‰, if If the value is ‰, the refitting process will be triggered (adjusting the initial parameters or expanding the fitting window) or the application will be transferred to manual review.

[0115] In practical implementation, the above system is integrated into a portable UBT detector. The overall dimensions are 25 cm × 18 cm × 10 cm, and the weight does not exceed 2.5 kg. Power is provided by a rechargeable lithium-ion battery with a capacity of 10,000 mAh, supporting continuous operation for over 8 hours. The human-machine interface is a 7-inch capacitive touchscreen, running a Linux-based real-time operating system (such as a custom kernel from the Yocto Project) to ensure deterministic task scheduling. The data storage unit uses 32 GB eMMC flash memory, supporting CSV format (for research analysis) and DICOM format (for hospital PACS systems) export. Communication interfaces include USB 3.0, Wi-Fi 6, and Bluetooth 5.2, meeting the data transmission needs of various scenarios.

[0116] In one specific embodiment, a 68-year-old male subject was diagnosed with Hp positivity via gastroscopy. His original signal... There was significant respiratory rhythm disturbance at baseline. The traditional three-point sampling method, due to single-point noise, results in a DOB of 3.8‰ (the critical value is usually 4.0‰), leading to false negatives. However, the system of this invention, through adaptive baseline correction and nonlinear fitting, calculates a DOB of 6.2‰, correctly identifying a positive result. (Fit residual...) ‰, energy concentration The quality control module determined it to be a valid sample.

[0117] In another comparative example, a 45-year-old female subject was diagnosed as Hp-negative by gastroscopy. Her original signal produced a sharp spurious peak (amplitude 4.5‰, duration <5 s) due to a brief cough. Traditional methods misidentified this spurious peak as a physiological peak, resulting in a DOB of 5.1‰ and a false positive result. The system of this invention successfully eliminated the spurious peak by constraining the respiratory rhythm (no reasonable metabolic response before and after the spurious peak) and verifying morphology (the second derivative does not conform to the characteristics of the metabolic curve), ultimately resulting in a DOB of 1.3‰ and a correct negative result.

[0118] See Figures 4-9 , Figures 4-9The user interface of the system described in this invention is shown. This invention achieves high-fidelity reconstruction of Helicobacter pylori breath test signals through a four-stage cascaded signal processing architecture. From hardware acquisition, digital filtering, baseline correction, peak identification to nonlinear fitting and DOB calculation, each step is meticulously designed based on physiological mechanisms and engineering practice, and is supplemented by rigorous quality control. This solution significantly improves detection accuracy and robustness without increasing the number of sampling points or improving hardware resolution, and is particularly suitable for low signal-to-noise ratio populations such as the elderly and children, possessing broad clinical application value and industrialization prospects. Those skilled in the art can easily implement all the technical contents of this invention based on the above detailed description and conventional engineering knowledge.

[0119] Example 2: This example is a further optimization based on Example 1. In this example, the identification and competitive fitting of bimodal metabolic signals are addressed to solve the technical problem that some subjects have bimodal metabolic signals in the breath test due to delayed gastric emptying or heterogeneity of Helicobacter pylori strains, and the existing single-peak fitting method cannot handle them correctly, resulting in serious deviations in DOB value calculation.

[0120] The specific implementation process is as follows:

[0121] 1. Scene and Data Input:

[0122] Assuming that after signal acquisition, filtering, and baseline correction, a drift-free signal is obtained .

[0123] The peak recognition unit identified two candidate physiological response peaks on the signal: the first peak lie in Seconds, amplitude ‰; second peak lie in Seconds, amplitude ‰. Total acquisition time of raw signal. Second.

[0124] 2. Preliminary determination of the existence of bimodal peaks (corresponding to step B1):

[0125] Calculate the peak pair time interval: Seconds. Judgment. Does it meet the requirements? The conditions are met, and it is preliminarily determined to be a potential bimodal signal.

[0126] 3. Peak purity verification:

[0127] calculate Local signal-to-noise ratio in the vicinity The evaluation interval is determined as follows: Seconds; to eliminate the influence of the peak point itself on the fluctuation assessment, it is removed from this interval. seconds (i.e.) (seconds) of data; calculate the remaining data points corresponding to... The standard deviation of is denoted as . ‰; Finally, calculate .

[0128] calculate Local signal-to-noise ratio in the vicinity The evaluation interval is determined as follows: Seconds; then remove from that interval seconds (i.e.) (seconds) of data; then calculate the data corresponding to the remaining data points. The standard deviation of is denoted as . ‰; Finally, calculate .

[0129] Both values ​​are greater than the threshold of 2.0, and the peak purity is verified, thus entering the bimodal competitive fitting process.

[0130] 4. Construct and initialize the bimodal fusion model (corresponding to step B2) Determine the extended fitting window:

[0131] Seconds. Construct a bimodal competitive fitting function:

[0132] ;

[0133] Among them, the shared residual offset .

[0134] Initialization parameters:

[0135] Weight: , .

[0136] Peak parameters:

[0137] for , , Second, Second, ;

[0138] for , , Second, Second, ; .

[0139] 5. Collaborative fitting and model competition (corresponding to steps B3 and B4): The system simultaneously fits two models.

[0140] Single-peak model fitting: with Centered on seconds, in the window Within seconds, Perform single-correction gamma function Fitting.

[0141] Bimodal model fitting: in extended window Within seconds, Perform function Fit the data and apply constraints: Second, Second.

[0142] After fitting, the following results are assumed:

[0143] Sum of Squared Residuals of Unimodal Model ,parameter data points .

[0144] Sum of Squared Residuals of Bimodal Model ,parameter data points .

[0145] Model comparisons were performed based on the sum of squared residuals, calculating the Bayesian Information Criterion (BIC) values ​​for unimodal and bimodal models respectively. The BIC calculation formula used for comparison is as follows:

[0146] ;

[0147] in, The number of model parameters. The total number of data points. This is the sum of squared residuals. Substitute the specific values ​​to calculate:

[0148] .

[0149] .because Therefore, the bimodal model was selected.

[0150] 6. Calculation and Output of Bimodal DOB (corresponding to step B5) Use the selected bimodal fitting curve , find it in Let the global maximum point within the interval be denoted as . Second, ‰. Calculate the baseline period. Within seconds The average value, assumed to be ‰. Therefore, the final DOB value is: ‰.

[0151] In contrast, if a unimodal model is forced, the resulting DOB value might be: The figure is approximately 1‰, which is an underestimation of about 17%.

[0152] The bimodal competitive fitting mechanism described in this embodiment solves the fundamental problem of model mismatch in the original single-peak fitting method when facing complex metabolic dynamics (bimodal). It automatically identifies single / bimodal modes through a data-driven approach and selects the optimal fitting model, avoiding forced single-peak fitting of bimodal signals and fundamentally eliminating the resulting DOB systematic error (which is usually underestimated). Bimodal identification itself may indicate specific physiological or pathological states, such as abnormal gastric emptying function or infection by specific strains, providing potential additional information to clinicians beyond infection positivity / negativeness. Through peak purity verification and rigorous model selection based on information criteria, it effectively prevents the misclassification of noise fluctuations or unseparated spurious peaks as metabolic bimodals, ensuring the reliability of the mechanism's activation.

[0153] Furthermore, when the system selects the bimodal model through a bimodal competitive fitting mechanism... As a final result, the quality control module accordingly calculates the bimodal model within its extended fitting window. Root mean square error of fitting residuals within .like If the value is ‰, then the refitting or manual review process will be triggered to ensure that the reliability of the output results meets the preset quality standards even in complex bimodal cases.

[0154] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0155] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for noise reduction and fitting of Helicobacter pylori breath test signals, characterized in that, Includes the following steps: The time-series data of the ¹³CO₂ / ¹²CO₂ isotope ratio in the exhaled breath of subjects after drug administration were continuously collected using an infrared spectroscopy sensor to form the raw signal. ,in This represents a time variable starting from the moment the medication was taken; For the original signal Apply bandpass digital filtering to obtain a preliminary purified signal. ; Based on the preliminary purification signal Constructing an adaptive baseline model from low-frequency components And perform baseline correction operation to generate drift-free signal. ; In the de-drift signal A robust peak identification algorithm based on morphological constraints is executed to determine the peak time point of the physiological response. and their corresponding amplitudes ; With the peak time point Centered on, within a preset time window Internally, a nonlinear function with physiological prior constraints is used to de-drift signals. Perform local fitting to generate a fitted curve. The DOB value is calculated based on the fitted curve.

2. The method for noise reduction and fitting of Helicobacter pylori breath test signals according to claim 1, characterized in that, The infrared spectral sensor is a dual-channel non-dispersive infrared detector. The center wavelength of the first channel of the dual-channel non-dispersive infrared detector is set to 4.26 μm, which is used to detect the absorption intensity of ¹²CO₂. The center wavelength of the second channel was set to 4.35 μm to detect the ¹³CO₂ absorption intensity. The original signal Defined as a time series of δ¹³C values, its calculation formula is: ； in , The value is the ¹³C / ¹²C ratio of PDB carbonate, which is 0.0112372 according to international standards.

3. The method for noise reduction and fitting of Helicobacter pylori breath test signals according to claim 1, characterized in that, The bandpass digital filtering process employs a fifth-order Butterworth bandpass filter with a low cutoff frequency. The high cutoff frequency is 0.005Hz. The frequency is 0.5Hz; the filtering operation is performed on the original signal. Zero-phase forward-backward filtering is used to eliminate phase distortion.

4. The method for noise reduction and fitting of Helicobacter pylori breath test signals according to claim 1, characterized in that, The adaptive baseline model The construction process includes: For the initial purification signal Empirical mode decomposition is performed to obtain several eigenmode functions and a residual term. ; Calculate the center frequency of each IMF component and will satisfy all Hz IMF components and residuals Add them together to form the initial baseline estimate. ; right Apply sliding window mid-range filtering, window length The time is 120 seconds, and the final adaptive baseline model is obtained. .

5. The method for noise reduction and fitting of Helicobacter pylori breath test signals according to claim 1, characterized in that, The robust peak identification algorithm includes: Calculate the drift-free signal first derivative With the second derivative ; retain satisfaction The time interval of ‰; Search within this interval and The local maxima are used as candidate peaks; Introducing respiratory rhythm constraints: If the time interval between two adjacent candidate peaks is less than 8 seconds or greater than 120 seconds, the one with the smaller amplitude is discarded; finally, the peak with the largest amplitude among the remaining candidate peaks is selected as the physiological response peak, and its corresponding time is... The amplitude is .

6. The method for noise reduction and fitting of Helicobacter pylori breath test signals according to claim 5, characterized in that, The first derivative With the second derivative The discrete expression is calculated using the five-point central difference scheme: ； ； in s.

7. The method for noise reduction and fitting of Helicobacter pylori breath test signals according to claim 1, characterized in that, The nonlinear function is a modified gamma distribution function: ； in , s; s， ， ‰； The preset time window s, The fitting process employs the Levenberg-Marquardt nonlinear least squares algorithm, with the aforementioned physiological prior constraints applied.

8. The method for noise reduction and fitting of Helicobacter pylori breath test signals according to claim 1, characterized in that, The formula for calculating the DOB value is: ； in s; The integration operation uses the composite Simpson's rule, with a step size of s, in Numerical integration was performed on 301 points sampled at 1-second intervals within the interval.

9. A Helicobacter pylori breath test signal noise reduction and fitting system, used to perform the Helicobacter pylori breath test signal noise reduction and fitting method according to any one of claims 1-8, characterized in that, include: The signal acquisition module is used to acquire the time series of the ¹³CO2 / ¹²CO2 isotope ratio in the exhaled breath of the subjects; A digital filtering unit, connected to the signal acquisition module, is used to apply bandpass digital filtering to the original signal; A baseline correction unit, connected to the digital filtering unit, is used to construct an adaptive baseline model and perform baseline correction. A peak identification unit, connected to the baseline correction unit, is used to execute a robust peak identification algorithm; A nonlinear fitting unit, connected to the peak identification unit, is used to perform local fitting of the drift-free signal within a preset time window; The DOB calculation unit, connected to the nonlinear fitting unit, is used to calculate the DOB value based on the fitted curve; the central processing unit schedules the above units uniformly through the internal bus.

10. The Helicobacter pylori breath test signal noise reduction and fitting system according to claim 9, characterized in that, The signal acquisition module includes a dual-channel NDIR sensor, a temperature compensation circuit, a pressure sensor, and an analog-to-digital converter. The dual-channel NDIR sensor has an optical cavity length of 15cm, uses a pulse-modulated infrared LED as its light source, and has a half-width at half-maximum (WHM) of 50nm for both channels. The analog-to-digital converter is a 24-bit Σ-Δ ADC, which is downsampled to 10Hz output via software. The temperature compensation circuit monitors the cavity temperature in real time using a platinum resistance thermometer (PT1000) and corrects the CO2 absorption coefficient for temperature based on the NIST standard gas database.

Images