A premature infant gastric juice reflux photoelectric sensing signal processing and early warning system

Abstract

This invention discloses a photoelectric sensor signal processing and early warning system for gastric reflux in premature infants, relating to the field of diagnostic analysis. It includes: a signal acquisition module for deploying photoelectric sensors and impedance sensors at corresponding body surface locations on the patient's stomach and esophagus, real-time acquisition of photoelectric reflection signals reflecting the presence and compositional characteristics of refluxed material in the esophagus, and simultaneous acquisition of impedance signals reflecting diaphragmatic activity; and an event recognition unit for analyzing signal characteristics based on the acquired photoelectric reflection signals using a preset recognition algorithm to determine whether a reflux event has occurred and to identify the type of refluxed material. This invention achieves non-invasive, continuous, and accurate monitoring of gastric reflux in premature infants, preliminarily distinguishes the nature of refluxed material through reflected light signal characteristics, and identifies the temporal and causal correlation patterns between reflux events and respiratory disturbances such as apnea and bradycardia, thereby enabling the prediction of respiratory complication risks.

Description

Technical Field

[0001] This invention relates to the field of diagnostic analysis technology, specifically to a photoelectric sensor signal processing and early warning system for gastric reflux in premature infants. Background Technology

[0002] Premature infants, especially those with extremely low birth weight, are at high risk of gastroesophageal reflux and related respiratory complications due to their immature gastrointestinal and respiratory systems. In the neonatal intensive care unit, reflux of gastric contents into the esophagus or even aspiration is a common trigger for serious events such as apnea, bradycardia, and decreased blood oxygen saturation, directly affecting the infant's survival rate and long-term neurodevelopment. Currently, routine monitoring of such events relies on vital sign monitoring such as heart rate, transcutaneous oxygen saturation, and respiratory plethysmography, while clinical judgment mainly depends on medical observation or invasive, intermittent examinations.

[0003] Current vital sign monitoring can only reflect secondary consequences of reflux, such as respiratory and circulatory problems, and cannot directly detect the primary event of reflux, resulting in a significant delay in early warning. Traditional monitoring cannot accurately correlate symptoms such as apnea and decreased blood oxygenation with specific reflux events in time, making it difficult to distinguish reflux-related respiratory events from other causative events, leading to a lack of targeted intervention. Visual observation by medical staff is not always reliable and cannot be continuous. Invasive methods such as esophageal pH monitoring can cause discomfort and infection risks to vulnerable premature infants, and can only reflect acid reflux, not non-acid reflux. Existing technologies are limited to displaying the current status or simple threshold alarms, lacking a risk assessment mechanism based on multimodal signal temporal pattern analysis. Therefore, there is an urgent need for a non-invasive, continuous system to monitor gastroesophageal reflux in premature infants and its correlation with respiratory events, in order to achieve early warning and precise intervention. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a photoelectric sensor signal processing and early warning system for gastric reflux in premature infants, which can effectively solve the problems of the existing technology.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] This invention discloses a photoelectric sensor signal processing and early warning system for gastric reflux in premature infants, comprising:

[0009] The signal acquisition module is used to collect photoelectric reflection signals reflecting the presence and composition characteristics of refluxed material in the esophagus in real time, and simultaneously collect impedance signals reflecting diaphragmatic activity, by attaching miniature photoelectric sensors and impedance sensors integrated on a flexible substrate in the form of button-shaped or strip-shaped sensing units to the corresponding body surface locations of the patient's stomach and esophagus. The miniature photoelectric sensor includes at least one light-emitting unit and one light-receiving unit. The light-emitting unit emits light signals of a specific wavelength, and the light-receiving unit receives the light signals reflected by the esophageal tissue and any refluxed material that may be present. Different components of refluxed material have different absorption and reflection characteristics for specific wavelengths of light.

[0010] The event recognition unit is used to analyze the signal characteristics based on the collected photoelectric reflection signal through a preset recognition algorithm, determine whether a reflux event has occurred, and identify the type of refluxed material, which includes milk, gastric juice, bile, or blood.

[0011] The status analysis module is used to convert impedance signals into diaphragmatic electrical signal data, analyze and define the patient's real-time respiratory status, and the reflux area of ​​refluxed material;

[0012] The interference identification module is used to construct an interference identification model based on deep learning algorithms. The model takes real-time identified regurgitation events and their types, as well as real-time respiratory status, as input. It combines several preset potential respiratory risk items to calculate and output the respiratory interference risk coefficient of the patient's respiratory risk item associated with the regurgitation event within a specific future time period. The several potential respiratory risk items include at least two of the following: risk of aspiration of regurgitated material, risk of regurgitation-induced apnea, risk of regurgitation-induced decrease in blood oxygen saturation, and risk of regurgitation-related bradycardia.

[0013] The data communication module is used to transmit refluxing events and their types, real-time respiratory status, and respiratory interference risk coefficients to external monitoring devices through a preset communication method; it adopts low-power Bluetooth or Zigbee wireless transmission protocol and has data encryption function; the external monitoring device is a dedicated monitor, tablet computer, or smartphone.

[0014] The alarm triggering module is used to automatically trigger a preset alarm when a regurgitation event is detected or the risk factor of respiratory interference exceeds a preset safety threshold, so as to prompt medical staff or caregivers to intervene.

[0015] Furthermore, the event recognition unit has sub-modules deployed at its lower level, including a feature extraction module, a classification decision module, and a model update module. The classification decision module is communicatively connected to the feature extraction module and the model update module, wherein:

[0016] The feature extraction module is used to extract feature vectors from the preprocessed photoelectric reflection signal, including time-domain features, frequency-domain features, and time-frequency-domain features, where:

[0017] Time-domain characteristics: mean signal amplitude, variance, zero-crossing rate, and baseline drift characteristics reflecting the presence of backflow material;

[0018] Frequency domain characteristics: The power spectrum of the signal is obtained by fast Fourier transform, and the energy proportion and dominant frequency of a specific frequency band are extracted. The default range of 0.1-2 Hz reflects fluid motion, and 2-10 Hz reflects the microscopic scattering characteristics of different components.

[0019] Time-frequency domain characteristics: By wavelet transform or empirical mode decomposition, the energy distribution of the signal in the time-frequency plane is obtained, which is used to characterize the suddenness of reflux events and different components, such as the fat particles in milk, the homogeneous liquid in gastric juice, and the unique scattering patterns brought by bile pigments in bile.

[0020] The classification decision module is used to pre-load and train a classification model through a convolutional neural network. The feature vector obtained from the feature extraction module is input into the classification model, and the classification model outputs the classification results regarding the reflux event and type. The classification results first determine whether there is reflux. When reflux is determined to be present, the classification result of the reflux substance type is output, including one or more of the following: milk, gastric juice, bile, and blood. Milk is characterized by a slowly varying envelope of signal amplitude and low-frequency energy dominating in the frequency domain; gastric juice is characterized by a relatively stable signal amplitude and a wide frequency domain energy distribution; bile is characterized by signal amplitude accompanied by specific high-frequency fluctuations and characteristic frequency band resonance in the time-frequency domain; blood is characterized by a unique bimodal or trimodal fluctuation pattern of signal amplitude, significant attenuation of light reflection signal in a specific hemoglobin absorption band, and a non-uniform scattered feature in the time-frequency distribution.

[0021] The model update module is used to incrementally learn or fine-tune the parameters of the classification model according to a preset scale based on the results of subsequent confirmed reflux events, so as to adapt to the physiological signal characteristics of individual patients.

[0022] Furthermore, the operating logic of the state analysis module is as follows:

[0023] The impedance signal is demodulated and converted to extract the electrical activity signal of the diaphragm that reflects the periodic contraction and relaxation of the diaphragm;

[0024] Waveform recognition and feature extraction were performed on the diaphragm electrical activity signal, and respiratory characteristic parameters, including respiratory cycle, inspiratory phase duration, expiratory phase duration, respiratory waveform amplitude, and waveform variability, were calculated.

[0025] The calculated real-time respiratory characteristic parameters are compared and analyzed with the preset normal respiratory parameter range thresholds for preterm infants of different gestational ages or weights.

[0026] Based on the results of comparative analysis, the child's real-time respiratory status is defined and output, including normal breathing, tachypnea, bradypnea, apnea, and irregular breathing.

[0027] Furthermore, the process of constructing the interference recognition model in the interference recognition module includes the following steps:

[0028] Step 41: Collect historical synchronously recorded esophageal photoelectroreflection signals, diaphragmatic impedance signals, clinically confirmed reflux events and their type labels, respiratory status labels, and labels of actual respiratory interference events in preterm infants; perform time alignment, segmentation, and labeling on the collected data to form a sample set containing input feature sequences and output label sequences;

[0029] Step 42: Extract the feature subsets related to regurgitation events and the feature subsets related to respiratory state from the input feature sequence of the sample set; the feature subsets related to regurgitation events include the duration, signal strength, type probability distribution and frequency of occurrence of regurgitation events; the feature subsets related to respiratory state include respiratory rate, rhythmicity index, inspiratory effort and respiratory waveform morphology;

[0030] Step 43: Construct a deep learning model architecture using a long short-term memory network, organize the feature subset into a temporal feature vector according to time windows, and use it as the model input; use the probability or risk level of whether a certain respiratory risk event will occur within a specific future time period as the model output; train the model using the sample set, and adjust the model parameters by optimizing the loss function, so that the model learns the temporal mapping relationship between the current reflux and respiratory features and the future respiratory risk;

[0031] Step 44: Evaluate the predictive performance metrics of the trained model using an independent validation dataset, including accuracy, recall, and specificity; iteratively optimize the model structure, hyperparameters, or input features based on the evaluation results until the model performance meets the preset clinical warning requirements.

[0032] Step 45: Integrate the final optimized model parameters into the interference identification model for online prediction of real-time input data and output the respiratory interference risk coefficient.

[0033] Furthermore, the formula for calculating the respiratory interference risk coefficient in the interference identification module is as follows:

[0034] ;

[0035] In the formula, The overall respiratory disturbance risk coefficient is a dimensionless scalar value. Represents the signal characteristics of backflow events The output value after processing with a non-linear activation function, which maps features to risk contribution values. This represents a multidimensional feature vector extracted from the photoelectric reflection signal and associated with the current backflow event. This represents the degree of deviation from the breathing state. ,in, Represents the first [number] extracted in real time from the diaphragm electrical signal. respiratory characteristic parameters, The first, representing the individualized or group standard for this child Normal reference values ​​for respiratory characteristic parameters The total number of respiratory characteristic parameters. Represents the total number of contextual risk factors. Represents the corresponding contextual risk factor The weight, Representing the Contextual risk factors, including but not limited to the infant's corrected gestational age, time after feeding, and body position, The time-series enhancement factor represents the type of backflow event. Density of historical risk events The function, , , and The weights representing different risk contributions are determined through optimization during the training process of the interference identification model.

[0036] Furthermore, the calculation process of the timing enhancement factor is as follows:

[0037] ;

[0038] In the formula, Representative and return type The relevant severity coefficient, Represents the past time window The number of respiratory disturbance events or high-risk warnings confirmed by the Inner Canon of Medicine. This represents the density influence coefficient.

[0039] Furthermore, during operation, the alarm triggering module sets up multiple alarm levels. When non-bile reflux is detected, the respiratory interference risk factor is in the low-risk range, or the reflux material location analysis indicates that the reflux material is mainly located in the lower esophagus, a first-level warning alarm is triggered. When bile reflux is detected, the respiratory interference risk factor enters the medium-high risk range, or the reflux material location analysis indicates that the reflux material has invaded the upper-middle esophagus or the vicinity of the pharynx, a second-level emergency alarm is triggered. When the reflux event occurs, the status analysis module determines that there is apnea, or the reflux material location analysis indicates that the reflux material has reached the pharynx and the airway inlet area, a third-level highest priority alarm is triggered.

[0040] Furthermore, the signal acquisition module is communicatively connected to a signal preprocessing module, which is used to filter, amplify, and denoise the acquired raw photoelectric reflection signal and impedance signal in order to extract effective signal components.

[0041] Furthermore, the data communication module is communicatively connected to a data storage module, which is used to store continuous raw signal data, processed feature data, identification event records, alarm logs and corresponding patient identification information, and supports data retrieval and export by time, event type or risk level.

[0042] Furthermore, the signal acquisition module is communicatively connected to the event recognition unit, the interference recognition module is communicatively connected to the event recognition unit, the status analysis module and the data communication module, and the data communication module is communicatively connected to the alarm triggering module.

[0043] (III) Beneficial Effects

[0044] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects:

[0045] 1. By non-invasively detecting changes in reflected light signals in the esophagus using photoelectric sensors applied to the body surface, the occurrence, intensity, and type of reflux events can be directly captured. Diaphragmatic impedance signals are used to non-invasively monitor respiratory effort and rhythm. This dual-modal synchronous monitoring enables non-invasive, continuous, and accurate monitoring of gastric reflux in premature infants. It can detect not only acidic and non-acidic reflux but also preliminarily distinguish the nature of refluxed material through reflected light signal characteristics, improving the comprehensiveness and specificity of reflux event identification. This provides detailed data for clinical practice, enabling medical staff to more accurately assess the actual frequency, duration, and composition of reflux.

[0046] 2. By deeply correlating and fusing regurgitation events with synchronously monitored respiratory states, and using a built-in deep learning interference identification model, the system can learn from complex physiological signals and identify the temporal and causal correlation patterns between regurgitation events and respiratory disturbances such as apnea and bradycardia. This enables the prediction of respiratory complication risks, allowing for earlier intervention windows in clinical treatment and effectively reducing the incidence of severe respiratory events. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0048] Figure 1 This is a schematic diagram of the overall framework of the present invention;

[0049] Figure 2 This is a schematic diagram of the event recognition unit in this invention;

[0050] Figure 3 This is a flowchart illustrating the process of constructing the interference identification model in this invention.

[0051] The labels in the diagram represent: 1. Signal acquisition module; 2. Event recognition unit; 21. Feature extraction module; 22. Classification decision module; 23. Model update module; 3. State analysis module; 4. Interference recognition module; 5. Data communication module; 6. Alarm triggering module; 7. Signal preprocessing module; 8. Data storage module. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0053] The present invention will be further described below with reference to embodiments.

[0054] Example 1

[0055] This embodiment describes a photoelectric sensor signal processing and early warning system for gastric reflux in premature infants, such as... Figures 1-3 As shown, it includes:

[0056] Signal acquisition module 1 is used to acquire photoelectric reflection signals reflecting the presence and composition characteristics of refluxed material in the esophagus in real time, and simultaneously acquire impedance signals reflecting diaphragmatic activity, by attaching miniature photoelectric sensors and impedance sensors integrated on a flexible substrate in the form of button-shaped or strip-shaped sensing units to the corresponding body surface locations of the patient's stomach and esophagus. The miniature photoelectric sensors include at least one light-emitting unit and one light-receiving unit. The light-emitting unit emits light signals of a specific wavelength, and the light-receiving unit receives the light signals reflected by esophageal tissue and any refluxed material present. Different components of refluxed material have different absorption and reflection characteristics for specific wavelengths of light. The logic for the light-emitting unit to emit light signals of a specific wavelength is as follows:

[0057] Based on the absorption spectral characteristics of the main light-absorbing components of gastric reflux, wavelengths with significant absorption peaks in the visible to near-infrared range were selected as candidate wavelengths. Considering the thickness, scattering coefficient, and absorption coefficient of the chest and abdominal wall tissues of preterm infants, a light propagation model in tissues was used for simulation. From the candidate wavelength set, a subset of wavelengths with a preset threshold tissue penetration depth and a high signal-to-noise ratio was selected. The degree of spectral overlap between the selected wavelength subset and common endogenous and exogenous interference sources, including changes in ambient light, motion artifacts, and fluctuations in skin blood flow, was analyzed. Wavelengths with low spectral overlap with interference sources and sensitivity to changes in reflux components were preferentially selected as the final selected wavelengths. Based on the final selected wavelengths, matching light-emitting diodes or laser diodes were selected as light sources, and driving circuits were configured to ensure stable operation at that wavelength. After system integration, the output light power and center wavelength of the light-emitting unit were calibrated using a standard reflector to ensure compliance with design specifications.

[0058] The signal acquisition module 1 is connected to the signal preprocessing module 7. The signal preprocessing module 7 is used to filter, amplify and denoise the acquired raw photoelectric reflection signal and impedance signal in order to extract the effective signal components.

[0059] Event recognition unit 2 is used to analyze signal characteristics based on the collected photoelectric reflection signal using a preset recognition algorithm to determine whether a reflux event has occurred and to identify the type of refluxed material, including milk, gastric juice, bile, or blood. Event recognition unit 2 has sub-modules, including a feature extraction module 21, a classification decision module 22, and a model update module 23. The classification decision module 22 is communicatively connected to the feature extraction module 21 and the model update module 23.

[0060] Feature extraction module 21 is used to extract feature vectors including time-domain features, frequency-domain features, and time-frequency-domain features from the preprocessed photoelectric reflection signal, wherein:

[0061] Time-domain characteristics: mean signal amplitude, variance, zero-crossing rate, and baseline drift characteristics reflecting the presence of backflow material;

[0062] Frequency domain characteristics: The power spectrum of the signal is obtained by fast Fourier transform, and the energy proportion and dominant frequency of a specific frequency band are extracted. The default range of 0.1-2 Hz reflects fluid motion, and 2-10 Hz reflects the microscopic scattering characteristics of different components.

[0063] Time-frequency domain characteristics: By wavelet transform or empirical mode decomposition, the energy distribution of the signal in the time-frequency plane is obtained, which is used to characterize the suddenness of reflux events and different components, such as the fat particles in milk, the homogeneous liquid in gastric juice, and the unique scattering patterns brought by bile pigments in bile.

[0064] The classification decision module 22 is used to pre-set and train a classification model through a convolutional neural network. The feature vector obtained by the feature extraction module 21 is input into the classification model, and the classification model outputs the classification results on the reflux event and type. The classification results first determine whether there is reflux. When it is determined that there is reflux, the classification result of the reflux material type is output, including one or more of milk, gastric juice, bile and blood. Milk is characterized by a slowly varying envelope of signal amplitude and low-frequency energy dominating in the frequency domain. Gastric juice is characterized by a relatively stable signal amplitude and a wide frequency domain energy distribution. Bile is characterized by signal amplitude accompanied by specific high-frequency fluctuations and characteristic frequency band resonance in the time and frequency domain. Blood is characterized by a unique bimodal or trimodal fluctuation pattern of signal amplitude, and significant attenuation of light reflection signal in a specific hemoglobin absorption band, with a non-uniform scattered feature in the time and frequency distribution.

[0065] The model update module 23 is used to incrementally learn or fine-tune the parameters of the classification model according to a preset scale based on the results of subsequent confirmed reflux events, so as to adapt to the physiological signal characteristics of individual patients.

[0066] Existing technologies often rely on single signal threshold judgments or simple time-domain analysis, making it difficult to distinguish complex reflux components and resulting in high false positive rates. This solution combines multidimensional signal feature extraction with an advanced ensemble classification model to achieve highly sensitive detection of reflux events. At the algorithmic level, it achieves accurate identification of four key types of substances: milk, gastric juice, bile, and blood. In particular, the identification of blood, by capturing its unique light absorption and scattering characteristics, provides crucial early warning capabilities for the early detection of esophageal mucosal damage or bleeding. Furthermore, the system possesses personalized learning capabilities, continuously optimizing identification accuracy and effectively overcoming the shortcomings of traditional methods in terms of poor adaptability and insufficient universality, providing a reliable and objective assessment tool for clinical use.

[0067] State analysis module 3 is used to convert impedance signals into diaphragmatic electrical signal data, analyze and define the patient's real-time respiratory status, and the reflux area of ​​refluxate; the operating logic of state analysis module 3 is as follows:

[0068] The impedance signal is demodulated and converted to extract the diaphragm electrical activity signal reflecting the periodic contraction and relaxation of the diaphragm. During this process, the acquired raw impedance signal is bandpass filtered to remove high-frequency noise and baseline drift, retaining the characteristic frequency bands reflecting diaphragm activity. From the preprocessed impedance signal, characteristic parameters related to the diaphragm contraction intensity and timing, such as signal amplitude envelope, zero-crossing rate, and energy proportion of specific frequency bands, are extracted. Based on the extracted characteristic parameters, a simulated diaphragm electrical signal waveform is generated using a predefined conversion function or lookup table. This simulated diaphragm electrical signal waveform is then smoothed, and the waveform amplitude and timing are dynamically calibrated according to individual differences or real-time signal quality to ensure consistency with physiological diaphragm electrical activity in key characteristics.

[0069] The transformation function or lookup table is obtained through training a machine learning model. The historical feature parameters of the machine learning model are used as input, and the real diaphragm electrical signal waveform acquired synchronously is used as the training target to learn the nonlinear mapping relationship from impedance features to electrical signal waveform.

[0070] Waveform recognition and feature extraction were performed on the diaphragm electrical activity signal, and respiratory characteristic parameters, including respiratory cycle, inspiratory phase duration, expiratory phase duration, respiratory waveform amplitude, and waveform variability, were calculated.

[0071] The calculated real-time respiratory characteristic parameters are compared and analyzed with the preset normal respiratory parameter range thresholds for preterm infants of different gestational ages or weights.

[0072] Based on the results of comparative analysis, the real-time respiratory status of the child is defined and output, including normal breathing, tachypnea, bradypnea, apnea, and irregular breathing.

[0073] Interference identification module 4 is used to construct an interference identification model based on deep learning algorithms. The model takes the real-time identified regurgitation events and their types, as well as the real-time respiratory status, as inputs. It combines several preset potential respiratory risk items to calculate and output the respiratory interference risk coefficient of the patient's respiratory risk item associated with the regurgitation event within a specific time period in the future. The several potential respiratory risk items include at least two of the following: risk of aspiration of regurgitated material, risk of regurgitation-induced apnea, risk of regurgitation-induced decrease in blood oxygen saturation, and risk of regurgitation-related bradycardia.

[0074] The data communication module 5 is used to transmit regurgitation events and their types, real-time respiratory status, and respiratory interference risk coefficients to external monitoring devices via a preset communication method; it adopts low-power Bluetooth or Zigbee wireless transmission protocols and has data encryption function; the external monitoring device is a dedicated monitor, tablet computer, or smartphone; the data communication module 5 is connected to the data storage module 8, which is used to store continuous raw signal data, processed feature data, identification event records, alarm logs, and corresponding patient identification information, and supports data retrieval and export by time, event type, or risk level.

[0075] The alarm triggering module 6 is used to automatically trigger a preset alarm when a reflux event is detected or the risk factor for respiratory disturbance exceeds a preset safety threshold, so as to prompt medical staff or caregivers to intervene. During operation, the alarm triggering module is set with multiple alarm levels. When non-biliary reflux is detected, the risk factor for respiratory disturbance is in the low-risk range, or the reflux material location analysis indicates that the reflux material is mainly located in the lower esophagus, a first-level warning alarm is triggered. When bile reflux is detected, the risk factor for respiratory disturbance enters the medium-high risk range, or the reflux material location analysis indicates that the reflux material has invaded the middle and upper esophagus or the vicinity of the pharynx, a second-level emergency alarm is triggered. When the conditions of a reflux event, the status analysis module's determination of apnea, or the reflux material location analysis indicating that the reflux material has reached the pharynx and airway inlet area are met simultaneously, a third-level highest priority alarm is triggered.

[0076] Signal acquisition module 1 is communicatively connected to event identification unit 2, interference identification module 4 is communicatively connected to event identification unit 2, status analysis module 3 and data communication module 5, and data communication module 5 is communicatively connected to alarm triggering module 6.

[0077] Compared with existing technologies, by simultaneously collecting and fusing esophageal photoelectric reflection signals and diaphragmatic impedance signals, this method effectively distinguishes between gastric reflux events and simple respiratory movements. It overcomes the technical bottlenecks of single signals being easily interfered with and having a high false alarm rate, thus improving the specificity and accuracy of reflux identification. Through a deep learning-based interference identification model, it can not only identify reflux that has already occurred, but also quantify the respiratory interference risk coefficient by analyzing reflux characteristics, breathing patterns, and clinical context information, thus buying valuable time for clinical intervention. The model can learn and adapt to the individualized physiological characteristics of different premature infants, and its risk calculation incorporates contextual factors such as the infant's gestational age and feeding status, making the early warning more targeted and clinically relevant, and effectively reducing unnecessary false alarms.

[0078] Example 2

[0079] At other levels, this embodiment also provides a process for constructing an interference recognition model, such as... Figure 2 As shown, it includes the following steps:

[0080] Step 41: Collect historical synchronously recorded esophageal photoelectroreflection signals, diaphragmatic impedance signals, clinically confirmed reflux events and their type labels, respiratory status labels, and labels of actual respiratory interference events in preterm infants; perform time alignment, segmentation, and labeling on the collected data to form a sample set containing input feature sequences and output label sequences;

[0081] Step 42: Extract the feature subsets related to regurgitation events and the feature subsets related to respiratory state from the input feature sequence of the sample set; the feature subsets related to regurgitation events include the duration, signal strength, type probability distribution and frequency of regurgitation events; the feature subsets related to respiratory state include respiratory rate, rhythmicity index, inspiratory effort and respiratory waveform morphology;

[0082] Step 43: Construct a deep learning model architecture using a long short-term memory network, organize the feature subset into a temporal feature vector according to time windows, and use it as the model input; use the probability or risk level of whether a certain respiratory risk event will occur within a specific future time period as the model output; train the model using a sample set, and adjust the model parameters by optimizing the loss function, so that the model learns the temporal mapping relationship between the current regurgitation and respiratory features and the future respiratory risk.

[0083] Step 44: Evaluate the predictive performance metrics of the trained model using an independent validation dataset, including accuracy, recall, and specificity; iteratively optimize the model structure, hyperparameters, or input features based on the evaluation results until the model performance meets the preset clinical warning requirements.

[0084] Step 45: Integrate the final optimized model parameters into the interference identification model for online prediction of real-time input data and output the respiratory interference risk coefficient.

[0085] The formula for calculating the risk factor of respiratory disturbance is:

[0086] ;

[0087] In the formula, The overall respiratory disturbance risk coefficient is a dimensionless scalar value. Represents the signal characteristics of backflow events The output value after processing with a non-linear activation function, which maps features to risk contribution values. This represents a multidimensional feature vector extracted from the photoelectric reflection signal and associated with the current backflow event. This represents the degree of deviation from the breathing state. ,in, Represents the first [unit / item] extracted in real time from the diaphragm electrical signal. respiratory characteristic parameters, The first, representing the individualized or group standard for this child Normal reference values ​​for respiratory characteristic parameters The total number of respiratory characteristic parameters. Represents the total number of contextual risk factors. Represents the corresponding contextual risk factor The weight, Representing the Contextual risk factors, including but not limited to the infant's corrected gestational age, time after feeding, and infant position, The time-series enhancement factor represents the type of backflow event. Density of historical risk events The calculation process for the timing enhancement factor is as follows:

[0088] ;

[0089] In the formula, Representative and return type The relevant severity coefficient, Represents the past time window The number of respiratory disturbance events or high-risk warnings confirmed by the Inner Canon of Medicine. Representative density influence coefficient;

[0090] , , and The weights representing different risk contributions are determined through optimization during the training process of the interference identification model.

[0091] This formula assesses the direct risk of the current regurgitation event and simultaneously evaluates the immediate physiological impact of the regurgitation event on the respiratory system. It incorporates the individual patient's condition and current care status as a risk modulator or amplifier. By including the risk pattern in the time dimension, if risk events or warnings have occurred frequently recently, or if the current regurgitation is of a high-risk type, this factor will significantly amplify the final risk coefficient.

[0092] By integrating the intensity of regurgitation events, real-time respiratory physiological deviation, clinical context, and event timeline patterns, and introducing model-optimized dynamic weights, a more comprehensive and clinically pathophysiological quantitative assessment of risk is achieved. Through the introduction of timeline enhancement factors, the system can capture clinical experience that recent high-frequency events predict higher risks, enabling it to identify dynamic risk accumulation effects, improving the accuracy of early warnings, and reducing missed or false alarms caused by isolated judgments.

[0093] In summary, this invention achieves continuous monitoring of gastric reflux events and their physiological effects by integrating photoelectric reflection and diaphragmatic impedance dual-path sensing. It can capture gastric reflux events with high sensitivity, avoiding the delay and invasive risks of traditional pH monitoring. Through the built-in interference identification model, it can dynamically analyze the temporal correlation between reflux characteristics and respiratory status. Combined with individualized physiological parameters and historical risk event density, it outputs a quantitative respiratory interference risk coefficient to achieve early warning. Furthermore, the model performance can be continuously improved through clinical data feedback to reduce the false alarm rate.

[0094] This invention combines real-time performance, accuracy, and clinical applicability, and can assist medical staff in timely intervention, reducing the risk of complications such as apnea and aspiration pneumonia caused by reflux in premature infants, and improving the quality of intensive care.

[0095] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A photoelectric sensor signal processing and early warning system for gastric reflux in premature infants, characterized in that, include: The signal acquisition module is used to deploy photoelectric sensors and impedance sensors to the corresponding body surface locations of the patient's stomach and esophagus, to collect photoelectric reflection signals that reflect the presence and composition characteristics of refluxed material in the esophagus in real time, and to collect impedance signals that reflect the activity of the diaphragm simultaneously. The event recognition unit is used to analyze the signal characteristics based on the collected photoelectric reflection signal through a preset recognition algorithm, determine whether a backflow event has occurred, and identify the type of backflow object. The status analysis module is used to convert impedance signals into diaphragmatic electrical signal data, analyze and define the patient's real-time respiratory status, and the reflux area of ​​refluxed material; The interference identification module is used to build an interference identification model based on deep learning algorithms. The model takes real-time identified regurgitation events and their types, as well as real-time respiratory status, as inputs. It combines several preset potential respiratory risk items to calculate and output the respiratory interference risk coefficient of the patient in a specific future time period that is associated with a certain respiratory risk item. The data communication module is used to transmit regurgitation events and their types, real-time respiratory status, and respiratory interference risk coefficients to external monitoring equipment through a preset communication method; The alarm triggering module is used to automatically trigger a preset type of alarm when a backflow event is detected or the risk factor of breathing interference exceeds a preset safety threshold.

2. The photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, The event recognition unit has sub-modules deployed below it. These sub-modules include a feature extraction module, a classification decision module, and a model update module. The classification decision module is communicatively connected to the feature extraction module and the model update module. The feature extraction module is used to extract feature vectors, including time-domain features, frequency-domain features, and time-frequency-domain features, from the preprocessed photoelectric reflection signal; The classification decision module is used to pre-set and train a classification model through a convolutional neural network. The feature vector obtained by the feature extraction module is input into the classification model, and the classification model outputs the classification results about the reflux event and type. The classification results first determine whether there is reflux. When it is determined that there is reflux, the classification result of the reflux material type is output, including one or more of the following: milk, gastric juice, bile and blood. The model update module is used to incrementally learn or fine-tune the parameters of the classification model according to a preset scale based on the results of subsequent confirmed backflow events.

3. The photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, The operating logic of the state analysis module is as follows: The impedance signal is demodulated and converted to extract the electrical activity signal of the diaphragm that reflects the periodic contraction and relaxation of the diaphragm; Waveform recognition and feature extraction were performed on the diaphragm electrical activity signal, and respiratory characteristic parameters, including respiratory cycle, inspiratory phase duration, expiratory phase duration, respiratory waveform amplitude, and waveform variability, were calculated. The calculated real-time respiratory characteristic parameters are compared and analyzed with the preset normal respiratory parameter range thresholds for preterm infants of different gestational ages or weights. Based on the results of comparative analysis, the child's real-time respiratory status is defined and output, including normal breathing, tachypnea, bradypnea, apnea, and irregular breathing.

4. The photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, The process of constructing the interference identification model in the interference identification module includes the following steps: Step 41: Collect historical synchronously recorded esophageal photoelectroreflection signals, diaphragmatic impedance signals, clinically confirmed reflux events and their type labels, respiratory status labels, and labels of actual respiratory interference events in preterm infants; perform time alignment, segmentation, and labeling on the collected data to form a sample set containing input feature sequences and output label sequences; Step 42: Extract the feature subsets related to the refluxing event and the feature subsets related to the respiratory state from the input feature sequence of the sample set; Step 43: Construct a deep learning model architecture using a long short-term memory network, organize the feature subset into a temporal feature vector according to time windows, and use it as the model input; use the probability or risk level of whether a certain respiratory risk event will occur within a specific future time period as the model output; Step 44: Use an independent validation dataset to evaluate the predictive performance metrics of the trained model, and iteratively optimize the model structure, hyperparameters, or input features based on the evaluation results until the model performance meets the preset clinical warning requirements. Step 45: Integrate the final optimized model parameters into the interference identification model and output the breathing interference risk coefficient.

5. The photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, The formula for calculating the respiratory interference risk coefficient in the interference identification module is as follows: ； In the formula, Represents the overall risk factor for respiratory disturbances. Represents the signal characteristics of backflow events The output value after processing with a non-linear activation function. This represents a multidimensional feature vector extracted from the photoelectric reflection signal and associated with the current backflow event. This represents the degree of deviation from the breathing state. The total number of contextual risk factors. Represents the corresponding contextual risk factor The weight, Representing the Item context risk factors, Represents the time-series enhancement factor. , , and The weights represent the different risk contribution items.

6. The photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 5, characterized in that, The calculation process of the time-series enhancement factor is as follows: ； In the formula, Representative and return type The relevant severity coefficient, Represents the past time window The number of respiratory disturbance events or high-risk warnings confirmed in the Inner Canon of Medicine. This represents the density influence coefficient.

7. The photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, During operation, the alarm triggering module is equipped with multiple alarm levels. When non-bile reflux is detected, the risk factor for respiratory interference is in the low-risk range, or the reflux material location analysis indicates that the reflux material is mainly located in the lower esophagus, a first-level warning alarm is triggered. When bile reflux is detected, the risk factor for respiratory interference enters the medium-high risk range, or the reflux material location analysis indicates that the reflux material has invaded the upper-middle esophagus or the vicinity of the pharynx, a second-level emergency alarm is triggered. When the following conditions are met simultaneously: a reflux event occurs, the status analysis module determines that there is apnea, or the reflux material location analysis indicates that the reflux material has reached the pharynx and airway inlet area, a Level 3 alarm with the highest priority is triggered.

8. The photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, The signal acquisition module is communicatively connected to a signal preprocessing module, which is used to filter, amplify, and denoise the acquired raw photoelectric reflection signal and impedance signal.

9. A photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, The data communication module is connected to a data storage module, which stores continuous raw signal data, processed feature data, identification event records, alarm logs and corresponding patient identification information, and supports data retrieval and export by time, event type or risk level.

10. A photoelectric sensor signal processing and early warning system for gastric reflux in premature infants according to claim 1, characterized in that, The signal acquisition module is communicatively connected to the event recognition unit, the interference recognition module is communicatively connected to the event recognition unit, the status analysis module and the data communication module, and the data communication module is communicatively connected to the alarm triggering module.

Images