Delivery assistant remote monitoring system and method

Abstract

The invention provides a midwife remote monitoring system and method, and the system comprises a data collection module which collects uterine contraction pressure waveform and fetal heart sound signals through a flexible sensing bellyband, obtains blood oxygen saturation and pulse wave conduction time signals through a physical sign monitoring ring, analyzes urine biochemical indexes through an intelligent analysis closestool, and a data processing module. The denoising module is used for denoising uterine contraction and fetal heart signals by adopting a wavelet threshold denoising algorithm, aligning uterine contraction waveforms and calculating a uterine contraction interval time sequence based on an improved dynamic time warping algorithm, and fusing multi-source data through an LSTM neural network to generate a physical sign parameter predicted value; a comprehensive risk score of a uterine contraction interval, a physical sign predicted value and a biochemical index is calculated based on a weighted evidence theory algorithm, a personalized physiological baseline model is constructed by utilizing Gaussian process regression, and a three-level early warning mechanism is realized to trigger emergency response step by step. According to the invention, real-time monitoring of the pregnant woman at home is realized through multi-modal sensing and an intelligent algorithm, and the detection rate of abnormal events is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical Internet of Things and remote monitoring technology, and specifically to a midwife remote monitoring system and method. Background Art

[0002] With the increasing demand for perinatal health management, achieving real-time and accurate monitoring of pregnant women's physiological parameters has become a key issue. Traditional remote monitoring solutions rely on static threshold alarm mechanisms, ignoring complex factors such as individual physiological differences among pregnant women, dynamic fluctuations in vital parameters, and the coupling relationship between multimodal data. Current technology suffers from three core flaws: signal interference. The uterine contraction pressure waves and fetal heart sounds collected by the abdominal band are easily interfered with by motion artifacts, leading to misjudgment; insufficient dynamic adaptability. The existing algorithm cannot automatically adjust the warning threshold according to the pregnant woman's circadian rhythm and body position changes; and delayed response. The system lacks the ability to predict complications, triggering alarms only after an abnormality occurs, delaying intervention.

[0003] In the existing technology, uterine contraction pressure is collected through a flexible sensor belly belt, blood oxygen saturation is monitored by a ring-type device, and biochemical indicators are analyzed by a smart toilet. Signal filtering and a simple rule engine are used to implement basic risk assessment. Although such solutions can partially improve data integrity, they still have significant limitations. The dynamic adaptability to the environment is weak. When the activity status of the pregnant woman suddenly changes or there is a sudden interference, the system cannot reconstruct the analysis model in real time, and the false alarm rate increases. There is also a passive response defect, which only issues an alarm when the current abnormality is detected, and preventive intervention capabilities are not established. Therefore, there is an urgent need to introduce dynamic modeling, predictive analysis and closed-loop control mechanisms to achieve breakthroughs in quality and efficiency. Summary of the Invention

[0004] In order to solve the technical problems mentioned in the current background technology, the purpose of the present invention is to provide a midwife remote monitoring system and method.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A midwife remote monitoring system and method, including:

[0007] Data acquisition module: Based on the flexible sensor belly belt, the system collects the pregnant woman's abdominal contraction pressure waveform and fetal heart sound signals in real time; based on the vital sign monitoring ring, the system collects the pregnant woman's blood oxygen saturation and pulse wave transmission time signals in real time; based on the intelligent analysis toilet, the system analyzes the pregnant woman's urine in real time to obtain biochemical indicators;

[0008] Data processing module: uses a wavelet threshold denoising algorithm to reduce noise on the contraction pressure waveform and fetal heart sound signal; aligns and calculates the contraction interval time series based on an improved dynamic time warping algorithm; inputs the contraction interval time series, blood oxygen saturation, and pulse wave transmission time signals into an LSTM neural network model to generate predicted values ​​for vital sign parameters;

[0009] Intelligent assessment module: calculates a comprehensive risk score based on the uterine contraction interval time series, predicted values ​​of physical sign parameters and biochemical indicators; constructs a personalized physiological baseline model for pregnant women and outputs a personalized physiological baseline;

[0010] Layered warning module: realizes three-layer warning based on the contraction interval time series, physical sign parameter prediction value and comprehensive risk score.

[0011] Furthermore, the flexible sensing abdominal belt adopts an elastic fabric base and has a built-in pressure sensor array and a microphone;

[0012] The pressure sensor array collects the uterine contraction pressure waveform of the pregnant woman's abdomen in real time, and the microphone captures the fetal heart sound signal, while adopting the adaptive filtering technology to suppress the environmental noise;

[0013] The vital sign monitoring ring has a built-in photoplethysmography sensor and a three-axis accelerometer;

[0014] The photoplethysmography sensor monitors the blood oxygen saturation and pulse waveform characteristics of pregnant women in real time;

[0015] Calculating the pulse wave transit time signal based on the pulse wave waveform characteristics;

[0016] The three-axis accelerometer collects motion signals and eliminates the interference of limb movement on PPG signals through a dynamic compensation algorithm;

[0017] The intelligent analysis toilet has a built-in multi-spectral urine analysis unit and a micro-robotic arm;

[0018] The micro-robotic arm automatically collects urine samples and drips them onto a test strip;

[0019] The analysis unit scans the reaction area of ​​the test paper to obtain biochemical indicators including protein, glucose, and fat.

[0020] Furthermore, the specific steps of the noise reduction process are as follows:

[0021] 1) Decomposing the uterine contraction pressure waveform and the fetal heart sound signal in the time-frequency domain based on the Symlet wavelet basis function to generate multi-scale wavelet coefficients W j,k , the formula is as follows:

[0022]

[0023] Among them, W j,k is the wavelet transform coefficient of the uterine contraction pressure waveform and the fetal heart sound signal at scale j and translation position k, x(t) is the original uterine contraction pressure waveform or the fetal heart sound signal, t is a continuous time variable, ψ j,k(t) is the Symlet wavelet basis function, j is the decomposition scale, and k is the translation parameter;

[0024] 2) Based on the multi-scale wavelet coefficient W j,k Perform scale-based noise suppression;

[0025] For the uterine contraction pressure waveform, the noise statistics of the pregnant woman in the resting state are used to dynamically adjust the threshold values ​​λ of each scale. j , in order to retain the effective physiological components;

[0026] Threshold adjustment formula:

[0027]

[0028] Among them, λ j is the dynamic threshold of the j-th scale, σ j is the j-th scale noise standard deviation estimated based on the resting state noise of pregnant women, and D is the signal length;

[0029] W j,k >λ j : is the effective uterine contraction pressure waveform;

[0030] For the fetal heart sound signal, the interference of environmental noise on the extraction of fetal heart rate features is eliminated by adaptive filtering technology;

[0031] The fetal heart sound signal is used to monitor the fetal heart rate in real time.

[0032] Furthermore, the uterine contraction interval time series is obtained based on the uterine contraction pressure waveform, and the specific steps are as follows:

[0033] 1) Aligning uterine contraction pressure waveforms using an improved dynamic time warping algorithm, including morphological similarity constraints and dynamic programming;

[0034] Real-time acquisition of the current uterine contraction pressure waveform sequence Q = {q1, q2, ..., q m} and the historical waveform sequence C={c1,c2,…,c n};

[0035] The dynamic programming aligns the waveform points of Q and C, accurately matching the starting points and peak points of adjacent contraction waveforms. The morphological similarity constraint optimizes the path search. The formula is as follows:

[0036]

[0037] Where C(a,b) is the aligned uterine contraction waveform sequence Q = {q1,q2,…,q m} and C={c1,c2,…,c n}, d(q a ,cb ) is the Euclidean distance, α is the diagonal path weight, and β is the horizontal and vertical path penalty weight;

[0038] 2) Extract the peak timestamps h of consecutive contraction events based on the aligned contraction pressure waveforms f , and calculate the uterine contraction interval time series H, the formula is as follows:

[0039] H f =h f -h f-1

[0040] Among them, H f is the fth element of the contraction interval time series.

[0041] Furthermore, the LSTM neural network model includes an input layer, a processing layer and an output layer;

[0042] The contraction interval time series, blood oxygen saturation and pulse wave transmission time signals are input into the input layer as physiological parameters. The input layer sets a time window of fixed length. The system continuously constructs the latest physiological parameters in the time window into a vector Y with a sliding step size. (t) ,

[0043]

[0044] Where N is the number of samples in the time window, H f (t) is the time series of uterine contraction intervals at time t, SpO2(t) is the blood oxygen saturation at time t, and PTT(t) is the pulse wave transmission time signal.

[0045] The processing layer captures long-term and short-term dependencies through a gating mechanism, including a forget gate, an input gate, and an output gate;

[0046] The forget gate retains valid historical information;

[0047] The historical information refers to the physiological parameters previously input;

[0048] The input gate filters the current valid historical information and integrates the newly input physiological parameters and valid historical information;

[0049] The output gate generates the predicted value of the vital sign parameter;

[0050] The output layer outputs the predicted value of the vital sign parameter.

[0051] Furthermore, the comprehensive risk score is calculated using a weighted evidence theory algorithm, including an evidence source setting module, a weight dynamic adjustment module, and a calculation module;

[0052] The evidence source setting module uses the uterine contraction interval time series, the predicted value of the physical sign parameter, and each of the biochemical indicators as evidence sources, sets a first weight for each evidence source based on historical clinical data and expert experience, and calculates the evidence source support and evidence source conflict degree;

[0053] The formula for calculating the support of evidence sources is as follows:

[0054]

[0055] Among them, S m is the support of the mth evidence source, e is the preset adjustment factor, x m is the actual value of the mth evidence source, μ m is the mean of the historical actual values ​​of the mth evidence source, σ m is the standard deviation of the historical actual value of the mth evidence source;

[0056] The conflict degree of evidence source is calculated based on the time window. The formula is as follows:

[0057]

[0058] Among them, C mn is the conflict degree between the mth evidence source and the nth evidence source, T is the length of the time window, is the actual value of the mth evidence source at time t, is the actual value of the nth evidence source at time t.

[0059] Furthermore, the weight dynamic adjustment module dynamically adjusts the first weight based on the first weight, the support degree of the evidence source and the conflict degree of the evidence source to obtain a second weight.

[0060]

[0061] Among them, w i ′ is the second weight of the mth evidence source, w m The first weight of the mth evidence source, N is the total number of evidence sources, is the preset conflict penalty coefficient, is the average conflict degree between the mth evidence source and other evidence sources;

[0062] The calculation module calculates the comprehensive risk score based on the evidence source support and the second weight, using the following formula:

[0063]

[0064] Among them, R c A score is assigned to the comprehensive risk.

[0065] Furthermore, the personalized physiological baseline model is constructed by a Gaussian process regression algorithm, and the uterine contraction interval time series, the predicted values ​​of the physical sign parameters and the biochemical indicators are used as a training set.

[0066] Furthermore, the three-level warning includes a first-level warning, a second-level warning, and a third-level warning;

[0067] The first level of warning is the abnormal warning of the uterine contraction interval time series;

[0068] The second-level warning is a warning of the degree of deviation between the predicted value of the individual characteristic parameter and the personalized physiological baseline;

[0069] The third level of early warning is to issue an early warning based on the comprehensive risk score;

[0070] Based on different levels of warning, the system activates the corresponding preset plan.

[0071] A midwife remote monitoring method, characterized by comprising:

[0072] Step 1: Using a flexible sensor belly belt to collect the pregnant woman's abdominal contraction pressure waveform and fetal heart sound signal in real time; using a vital sign monitoring ring to collect the pregnant woman's blood oxygen saturation and pulse wave transmission time signal in real time; using an intelligent analysis toilet to analyze the pregnant woman's urine in real time to obtain biochemical indicators;

[0073] Step 2: Using a wavelet threshold denoising algorithm to reduce noise on the contraction pressure waveform and fetal heart sound signal; aligning and calculating the contraction interval time series based on an improved dynamic time warping algorithm; inputting the contraction interval time series, blood oxygen saturation, and pulse wave transmission time signals into an LSTM neural network model to generate predicted values ​​of vital sign parameters;

[0074] Step 3: Calculate a comprehensive risk score based on the uterine contraction interval time series, predicted values ​​of physical sign parameters, and biochemical indicators; construct a personalized physiological baseline model for pregnant women, and output a personalized physiological baseline;

[0075] Step 4: Implement a three-level early warning based on the uterine contraction interval time series, the predicted value of the physical sign parameter and the comprehensive risk score.

[0076] Compared with the prior art, the advantages of the present invention are:

[0077] 1. This invention uses an LSTM neural network to fuse multiple physiological parameters, such as uterine contraction interval, blood oxygen saturation, and pulse wave transmission time, in real time to generate physical sign prediction values. Combined with urine biochemical indicators, this method breaks through the limitations of traditional single-parameter static analysis. This deep collaboration significantly improves the ability to predict abnormal events early, achieving a shift from passive response to active prediction.

[0078] 2. This invention uses Gaussian process regression to construct a personalized physiological baseline model for pregnant women. It combines this with a weighted evidence theory algorithm to dynamically adjust the weights of multiple evidence sources, addressing the problem that traditional static thresholds cannot adapt to individual physiological fluctuations. The system can autonomously optimize the risk assessment model as the pregnant woman's condition changes, significantly reducing the false alarm rate.

[0079] 3. The present invention designs a hierarchical early warning module, which triggers the response strategy step by step according to abnormal uterine contraction intervals, deviation of physical signs from personalized baselines, and comprehensive risk scores. Each level of early warning is matched with a differentiated treatment plan to form a "monitoring-assessment-early warning-intervention" closed-loop management, effectively shortening the emergency response delay. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0081] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0082] Figure 2 This is a schematic diagram of the comprehensive risk score calculation process of the present invention;

[0083] Figure 3 Schematic diagram of the data acquisition module of the present invention. DETAILED DESCRIPTION

[0084] To achieve the above objectives, the present invention is implemented through the following technical solutions. The present invention provides a midwife remote monitoring system and method, please refer to Figures 1 to 3 , the system comprises:

[0085] The data acquisition module collects the abdominal contraction pressure waveform and fetal heart sound signal of pregnant women in real time based on the flexible sensing belly belt; collects the blood oxygen saturation and pulse wave conduction time signal of pregnant women in real time based on the vital sign monitoring ring; and analyzes the urine of pregnant women in real time based on the intelligent analysis toilet to obtain an analysis report.

[0086] The flexible sensing belly belt uses an elastic fabric base with an integrated pressure sensor array and a micro-microphone. It fits snugly around the abdomen of pregnant women, ensuring comfort and stable signal acquisition.

[0087] The flexible sensing belly belt collects the uterine contraction pressure waveform of the pregnant woman's abdomen in real time through the pressure sensor array. The sampling frequency is not less than 50Hz and the dynamic range covers 0-100mmHg.

[0088] The flexible sensor belly belt uses a micro-microphone to capture fetal heart sound signals, combined with adaptive filtering technology to suppress environmental noise and extract fetal heart sound signals;

[0089] The vital sign monitoring ring has a built-in photoplethysmography (PPG) sensor and a three-axis accelerometer, and is made of medical-grade silicone material, allowing for seamless wear around the clock.

[0090] The vital sign monitoring ring uses a photoplethysmography (PPG) sensor to monitor the pregnant woman's blood oxygen saturation (SpO2) and pulse waveform characteristics in real time, with a measurement range of 70%-100% and an accuracy of ±2%;

[0091] The pulse wave transit time (PTT) signal is calculated based on the pulse wave waveform characteristics to indirectly evaluate the blood pressure change trend;

[0092] The vital sign monitoring ring uses a three-axis accelerometer to synchronously collect motion signals, and combines a dynamic compensation algorithm to eliminate the interference of limb movement on the PPG signal.

[0093] The intelligent analysis toilet has a built-in multi-spectral urine analysis unit and a micro-robotic arm, which supports automatic sampling and test strip dipping, and detects color changes of the test strip through an optical sensor;

[0094] After a pregnant woman uses the intelligent analysis toilet, the robotic arm automatically collects a urine sample and drips it onto a test strip. The analysis unit then scans the reaction area of ​​the test strip to obtain biochemical indicators including protein, glucose, and fat.

[0095] The data processing module uses a wavelet threshold denoising algorithm to reduce the noise of the contraction pressure waveform and fetal heart sound signal; aligns and calculates the contraction interval time series based on an improved dynamic time warping algorithm; inputs the contraction interval time series and the blood oxygen saturation and pulse wave transmission time signals into the LSTM neural network to generate the predicted value of the vital sign parameters.

[0096] The wavelet threshold denoising algorithm is used to optimize the uterine contraction pressure waveform (sampling frequency ≥ 50 Hz, dynamic range 0-100 mmHg) and fetal heart sound signal collected by the flexible sensor belly belt:

[0097] Fetal heart sound signals are used to monitor fetal heart rate (FHR) in real time;

[0098] The uterine contraction pressure waveform and fetal heart sound signal are decomposed at multiple scales using the Symlet wavelet basis function. The Symlet wavelet basis function is:

[0099]

[0100] Among them, D j,kis the wavelet transform coefficient of the uterine contraction pressure waveform and fetal heart sound signal at scale j and translation position k, x(t) is the original uterine contraction pressure waveform or fetal heart sound signal, t is a continuous time variable, ψ j,k (t) is the Symlet wavelet basis function, j is the decomposition scale, and k is the translation parameter.

[0101] For the baseline noise characteristics of the uterine contraction pressure waveform in the resting state of pregnant women, the thresholds of each scale are dynamically adjusted to retain the effective physiological components;

[0102] Threshold adjustment formula:

[0103]

[0104] Among them, λ j is the dynamic threshold of the j-th scale, σ j is the j-th scale noise standard deviation estimated based on the resting state noise of pregnant women, and D is the signal length;

[0105] W j,k >λ j : is the effective uterine contraction pressure waveform;

[0106] For fetal heart sound signals, adaptive filtering technology is used to eliminate the interference of environmental noise on fetal heart rate feature extraction, ensuring that the signal-to-noise ratio is ≥30dB.

[0107] The uterine contraction interval time series is obtained based on the uterine contraction pressure waveform. The specific steps are as follows

[0108] 1) An improved dynamic time warping (DTW) algorithm was used to align the uterine contraction pressure waveforms, including morphological similarity constraints and dynamic programming. The specific steps are as follows;

[0109] The real-time collected uterine contraction pressure waveform sequence Q={q1,q2,…,q m}(current cycle) and historical waveform sequence C={c1,c2,…,c n},

[0110] The morphological similarity constraint optimization path search is introduced. The waveform points of Q and C are aligned through dynamic programming to accurately match the starting points and peak points of adjacent contraction waveforms. The formula is as follows:

[0111]

[0112] Where C(a,b) is the aligned uterine contraction waveform sequence Q = {q1,q2,…,q m} and C={c1,c2,…,c n}, d(q a ,c b ) is the Euclidean distance, which measures the distance between two waveform points qa (current contraction) and c b The morphological difference of the reference contraction is calculated, where α is the diagonal path weight (default 1.0) and β is the horizontal / vertical path penalty weight (default 1.5), which is optimized based on historical contraction data.

[0113] 2) Extract the peak timestamps h of consecutive contraction events based on the aligned contraction pressure waveforms f ,

[0114] Calculate the time series H between contractions f , the calculation formula is as follows:

[0115] H f =h f -h f-1

[0116] Among them, H f is the time series of the interval between the fth contraction, is the peak timestamp of the f-th uterine contraction pressure waveform;

[0117] H f <2 minutes and H f Waveforms >10 minutes are marked as abnormal.

[0118] The uterine contraction interval time series, blood oxygen saturation and pulse wave transmission time signals are input into the preset LSTM neural network model as physiological parameters;

[0119] The LSTM neural network model consists of input layer, processing layer and output layer.

[0120] The input layer receives the time series of uterine contraction intervals, blood oxygen saturation, and pulse wave transmission time signals, and sets the time window;

[0121] In order to effectively capture the temporal dynamic characteristics of these physiological parameters and their mutual dependencies, the input layer sets a fixed-length time window (the default length is 60 seconds), and the system continuously constructs the latest physiological parameters in the time window into a vector Y with a sliding step (the default length is 10 seconds). (t) ,

[0122]

[0123] Where N is the number of samples in the time window, H f (t) is the time series of uterine contraction intervals at time t, SpO2(t) is the blood oxygen saturation at time t, and PTT(t) is the pulse wave transmission time signal.

[0124] The processing layer captures long-term and short-term dependencies through a gating mechanism, including a forget gate, input gate, and output gate;

[0125] The forget gate retains valid historical information:

[0126] f t =σ(Z f ·[V t-1 ,Y (t) ]+b a )

[0127] Among them, f t is the forget gate control vector at time t, 0 means completely discarding the corresponding historical information, 1 means completely retaining the corresponding historical information, (0,1) means retaining the historical information in proportion, σ is the Sigmoid function, Z f is the forget gate weight matrix, V t-1 is the hidden state of the previous time step, b a is the forget gate bias vector;

[0128] Historical information refers to previously entered information;

[0129] The input gate filters the currently valid historical information:

[0130]

[0131] Among them, g t is the input gate vector at time t, Z e With Z c are the input gate and candidate cell state weights, b b with b c are the bias vectors of the input gate and candidate state respectively; is the candidate cell state, tanh is the activation function;

[0132] Based on the results of the forget gate and input gate, the cell state is updated and the new and old information are integrated:

[0133]

[0134] Among them, ⊙ represents element-by-element multiplication;

[0135] The output gate generates the current hidden state:

[0136]

[0137] Among them, t is the output gate vector at time t, Z o is the output gate weight, b o is the output gate bias vector, V T is the current hidden state;

[0138] Prediction value generation stage:

[0139] y pred=Z y ·V T +b y

[0140] Among them, y pred is the predicted value vector, Z y is the output layer weight matrix, V T is the hidden state at the final time step, b y is the output layer bias vector;

[0141] The weight matrix and bias vector are obtained by training the LSTM neural network model;

[0142] The output layer generates predicted values ​​of vital sign parameters for the next 5 minutes (including uterine contraction frequency, SpO2 decline rate, and PTT coefficient of variation).

[0143] An intelligent assessment module calculates a comprehensive risk score based on the uterine contraction interval time series, predicted values ​​of physical sign parameters, and biochemical indicators; constructs a personalized physiological baseline model for pregnant women, and outputs a personalized physiological baseline;

[0144] The comprehensive risk score is calculated using a weighted evidence theory algorithm, which includes an evidence source setting module, a weight dynamic adjustment module, and a calculation module;

[0145] The evidence source setting module uses each of the contraction interval time series, predicted values ​​of physical sign parameters, and biochemical indicators as evidence sources, sets a first weight for each evidence source based on historical clinical data and expert experience, and calculates the evidence source support and evidence source conflict degree;

[0146] Evidence source support is used to quantify the credibility of each evidence source. The calculation formula is as follows:

[0147]

[0148] Among them, S m is the support of m evidence sources, e is the preset adjustment factor, x m is the actual value of the mth evidence source, μ m is the mean of the historical actual values ​​of the mth evidence source, σ m is the standard deviation of the historical actual value of the mth evidence source;

[0149] The evidence source conflict degree measures the degree of contradiction between different evidence sources and is calculated based on the time window. The formula is as follows:

[0150]

[0151] Among them, C mn is the conflict degree between the mth evidence source and the nth evidence source, T is the length of the time window (default T = 60 seconds), is the actual value of the nth evidence source at time t, is the actual value of the nth evidence source at time t;

[0152] The weight dynamic adjustment module dynamically adjusts the first weight based on the first weight, the support degree of the evidence source, and the conflict degree of the evidence source to obtain the second weight.

[0153]

[0154] Among them, w′ i is the second weight of the mth evidence source, w m The first weight of the mth evidence source, N is the total number of evidence sources, ∈ is the preset conflict penalty coefficient, is the average conflict degree between the mth evidence source and other evidence;

[0155] The calculation module calculates the comprehensive risk score R based on the support of the evidence source and the second weight c , the formula is as follows:

[0156]

[0157] A personalized physiological baseline model for pregnant women was constructed based on the Gaussian process regression algorithm, and the uterine contraction interval time series, predicted values ​​of physical sign parameters and biochemical indicators were used as training sets to train the personalized physiological baseline model.

[0158] A layered warning module realizes three-layer warning based on the uterine contraction interval time series, the predicted value of the physical sign parameters and the comprehensive risk score;

[0159] The three-tier warning includes the first-tier warning, the second-tier warning, and the third-tier warning;

[0160] The first level of warning uses the contraction interval time series to issue an abnormal warning, which is divided into three levels;

[0161] The first level of warning is when the interval between single contractions is less than 2 minutes or more than 10 minutes. The flexible sensor belly belt will vibrate at a low frequency and a text alert will be pushed to the pregnant woman's mobile terminal, suggesting that she adjust her body position or contact a midwife.

[0162] The second-level warning is when the interval between two consecutive contractions is abnormal. The medium-frequency vibration of the vital sign monitoring ring is activated, and the system generates a uterine contraction waveform and time series analysis chart. A pop-up window reminder is sent to the midwife, along with the uterine contraction waveform and time series analysis chart.

[0163] The third-level warning is when the intervals between three consecutive contractions are abnormal. The system triggers a high-frequency vibration alarm to notify the emergency center, and generates an abnormal contraction report and sends it to the emergency center simultaneously.

[0164] The second level of warning, the personalized physiological baseline model receives the predicted values ​​of physical sign parameters in real time, and analyzes the deviation degree between the predicted values ​​of physical sign parameters and the physiological baseline in real time, which is divided into three levels:

[0165] The first deviation will push a prompt message to the pregnant woman and her family members, suggesting that they strengthen observation;

[0166] The second deviation automatically triggers a pop-up window alert for the midwife, along with a trend curve and historical data comparison chart;

[0167] The third deviation system activates the emergency communication link, pushes a red alarm command to the hospital's obstetrics center, and simultaneously transmits the real-time vital signs data of the pregnant woman.

[0168] The third level of warning is divided into three levels of risk warning based on the comprehensive risk score.

[0169] The first risk warning pushes daily advice to pregnant women;

[0170] The second risk warning sends an alert to the midwife's terminal, suggesting a remote consultation;

[0171] The third risk warning triggers an emergency alarm, automatically connects to the emergency system and sends the pregnant woman's geographic location, electronic medical record summary and multimodal data snapshot.

[0172] All data in the system is encrypted, stored, and transmitted using lightweight blockchain technology (such as IOTA Tangle);

[0173] The personal information of pregnant women is converted into irreversible hash values.

[0174] A midwife remote monitoring method, characterized by comprising:

[0175] Step 1: Using a flexible sensor belly belt to collect the pregnant woman's abdominal contraction pressure waveform and fetal heart sound signal in real time; using a vital sign monitoring ring to collect the pregnant woman's blood oxygen saturation and pulse wave transmission time signal in real time; using an intelligent analysis toilet to analyze the pregnant woman's urine in real time to obtain biochemical indicators;

[0176] Step 2: Using a wavelet threshold denoising algorithm to reduce noise on the contraction pressure waveform and fetal heart sound signal; aligning and calculating the contraction interval time series based on an improved dynamic time warping algorithm; inputting the contraction interval time series, blood oxygen saturation, and pulse wave transmission time signals into an LSTM neural network model to generate predicted values ​​of vital sign parameters;

[0177] Step 3: Calculate a comprehensive risk score based on the uterine contraction interval time series, predicted values ​​of physical sign parameters, and biochemical indicators; construct a personalized physiological baseline model for pregnant women, and output a personalized physiological baseline;

[0178] Step 4: Implement a three-level early warning based on the uterine contraction interval time series, the predicted value of the physical sign parameter and the comprehensive risk score.

[0179] In summary, the present invention realizes seamless switching between hospital and home scenarios through the collaboration of micro wearable sensors and wireless transmission modules. Compared with traditional fixed monitoring equipment, it eliminates spatial constraints and ensures continuous monitoring of high-risk pregnant women during daily activities.

[0180] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A midwife remote monitoring system, characterized in that: include; Data acquisition module: Based on the flexible sensor belly belt, the system collects the pregnant woman's abdominal contraction pressure waveform and fetal heart sound signals in real time; based on the vital sign monitoring ring, the system collects the pregnant woman's blood oxygen saturation and pulse wave transmission time signals in real time; based on the intelligent analysis toilet, the system analyzes the pregnant woman's urine in real time to obtain biochemical indicators; Data processing module: uses a wavelet threshold denoising algorithm to reduce noise on the contraction pressure waveform and fetal heart sound signal; aligns and calculates the contraction interval time series based on an improved dynamic time warping algorithm; inputs the contraction interval time series, blood oxygen saturation, and pulse wave transmission time signals into an LSTM neural network model to generate predicted values ​​for vital sign parameters; Intelligent assessment module: calculating a comprehensive risk score based on the uterine contraction interval time series, predicted values ​​of physical sign parameters and biochemical indicators; Build a personalized physiological baseline model for pregnant women and output a personalized physiological baseline; Layered warning module: realizes three-layer warning based on the contraction interval time series, physical sign parameter prediction value and comprehensive risk score.

2. The system according to claim 1, wherein: The flexible sensing abdominal belt adopts an elastic fabric base and has a built-in pressure sensor array and microphone; The pressure sensor array collects the uterine contraction pressure waveform of the pregnant woman's abdomen in real time, and the microphone captures the fetal heart sound signal, while adopting the adaptive filtering technology to suppress the environmental noise; The vital sign monitoring ring has a built-in photoplethysmography sensor and a three-axis accelerometer; The photoplethysmography sensor monitors the blood oxygen saturation and pulse waveform characteristics of pregnant women in real time; Calculating the pulse wave transit time signal based on the pulse wave waveform characteristics; The three-axis accelerometer collects motion signals and eliminates the interference of limb movement on PPG signals through a dynamic compensation algorithm; The intelligent analysis toilet has a built-in multi-spectral urine analysis unit and a micro-robotic arm; The micro-robotic arm automatically collects urine samples and drips them onto a test strip; The analysis unit scans the reaction area of ​​the test paper to obtain biochemical indicators including protein, glucose, and fat.

3. The system according to claim 2, characterized in that The specific steps of the noise reduction process are as follows: 1) Decomposing the uterine contraction pressure waveform and the fetal heart sound signal in the time-frequency domain based on the Symlet wavelet basis function to generate multi-scale wavelet coefficients W j,k , the formula is as follows: Among them, W j,k is the wavelet transform coefficient of the uterine contraction pressure waveform and the fetal heart sound signal at scale j and translation position k, x(t) is the original uterine contraction pressure waveform or the fetal heart sound signal, t is a continuous time variable, ψ j,k (t) is the Symlet wavelet basis function, j is the decomposition scale, and k is the translation parameter; 2) Based on the multi-scale wavelet coefficient W j,k Perform scale-based noise suppression; For the uterine contraction pressure waveform, the noise statistics of the pregnant woman in the resting state are used to dynamically adjust the threshold values ​​λ of each scale. j , in order to retain the effective physiological components; Threshold adjustment formula: Among them, λ j is the dynamic threshold of the j-th scale, σ j is the j-th scale noise standard deviation estimated based on the resting state noise of pregnant women, and D is the signal length; W j,k >λ j : is the effective uterine contraction pressure waveform; For the fetal heart sound signal, the interference of environmental noise on the extraction of fetal heart rate features is eliminated by adaptive filtering technology; The fetal heart sound signal is used to monitor the fetal heart rate in real time.

4. The system according to claim 3, characterized in that The uterine contraction interval time series is obtained based on the uterine contraction pressure waveform, and the specific steps are as follows: 1) Aligning uterine contraction pressure waveforms using an improved dynamic time warping algorithm, including morphological similarity constraints and dynamic programming; Real-time acquisition of the current uterine contraction pressure waveform sequence Q = {q1, q2, ..., q m } and the historical waveform sequence C={c1,c2,…,c n }; The dynamic programming aligns the waveform points of Q and C, accurately matching the starting points and peak points of adjacent contraction waveforms. The morphological similarity constraint optimizes the path search. The formula is as follows: Where C(a,b) is the aligned uterine contraction waveform sequence Q = {q1,q2,…,q m } and C={c1,c2,…,c n }, d(q a ,c b ) is the Euclidean distance, α is the diagonal path weight, and β is the horizontal and vertical path penalty weight; 2) Extract the peak timestamps h of consecutive contraction events based on the aligned contraction pressure waveforms f , and calculate the uterine contraction interval time series H, the formula is as follows: H f =h f -h f-1 Among them, H f is the fth element of the contraction interval time series.

5. The system according to claim 1, wherein: The LSTM neural network model includes an input layer, a processing layer and an output layer; The contraction interval time series, blood oxygen saturation and pulse wave transmission time signals are input into the input layer as physiological parameters. The input layer sets a time window of fixed length. The system continuously constructs the latest physiological parameters in the time window into a vector Y with a sliding step size. (t) , Where N is the number of samples in the time window, H f (t) is the time series of uterine contraction intervals at time t, SpO2(t) is the blood oxygen saturation at time t, and PTT(t) is the pulse wave transmission time signal. The processing layer captures long-term and short-term dependencies through a gating mechanism, including a forget gate, an input gate, and an output gate; The forget gate retains valid historical information; The historical information refers to the physiological parameters previously input; The input gate filters the current valid historical information and integrates the newly input physiological parameters and valid historical information; The output gate generates the predicted value of the vital sign parameter; The output layer outputs the predicted value of the vital sign parameter.

6. The system according to claim 1, wherein: The comprehensive risk score is calculated using a weighted evidence theory algorithm, including an evidence source setting module, a weight dynamic adjustment module, and a calculation module; The evidence source setting module uses the uterine contraction interval time series, the predicted value of the physical sign parameter, and each of the biochemical indicators as evidence sources, sets a first weight for each evidence source based on historical clinical data and expert experience, and calculates the evidence source support and evidence source conflict degree; The formula for calculating the support of evidence sources is as follows: Among them, S m is the support of the mth evidence source, e is the preset adjustment factor, x m is the actual value of the mth evidence source, μ m is the mean of the historical actual values ​​of the mth evidence source, σ m is the standard deviation of the historical actual value of the mth evidence source; The conflict degree of evidence source is calculated based on the time window. The formula is as follows: Among them, C mn is the conflict degree between the mth evidence source and the nth evidence source, T is the length of the time window, is the actual value of the mth evidence source at time t, is the actual value of the nth evidence source at time t.

7. The system according to claim 6, characterized in that The weight dynamic adjustment module dynamically adjusts the first weight based on the first weight, the support degree of the evidence source and the conflict degree of the evidence source to obtain a second weight. Among them, w′ i is the second weight of the mth evidence source, w m The first weight of the mth evidence source, N is the total number of evidence sources, ∈ is the preset conflict penalty coefficient, is the average conflict degree between the mth evidence source and other evidence sources; The calculation module calculates the comprehensive risk score based on the evidence source support and the second weight, using the following formula: Among them, R c A score is assigned to the comprehensive risk.

8. The system according to claim 1, wherein: The personalized physiological baseline model is constructed by a Gaussian process regression algorithm, and the uterine contraction interval time series, the predicted values ​​of the physical sign parameters and the biochemical indicators are used as a training set.

9. The system according to claim 1, wherein: The three-tier warning includes the first-tier warning, the second-tier warning and the third-tier warning; The first level of warning is the abnormal warning of the uterine contraction interval time series; The second-level warning is a warning of the degree of deviation between the predicted value of the individual characteristic parameter and the personalized physiological baseline; The third level of early warning is to issue an early warning based on the comprehensive risk score; Based on different levels of warning, the system activates the corresponding preset plan.

10. A midwife remote monitoring method, characterized in that: include; Step 1: Using a flexible sensor belly belt to collect the pregnant woman's abdominal contraction pressure waveform and fetal heart sound signal in real time; using a vital sign monitoring ring to collect the pregnant woman's blood oxygen saturation and pulse wave transmission time signal in real time; using an intelligent analysis toilet to analyze the pregnant woman's urine in real time to obtain biochemical indicators; Step 2: Using a wavelet threshold denoising algorithm to reduce noise on the contraction pressure waveform and fetal heart sound signal; aligning and calculating the contraction interval time series based on an improved dynamic time warping algorithm; inputting the contraction interval time series, blood oxygen saturation, and pulse wave transmission time signals into an LSTM neural network model to generate predicted values ​​of vital sign parameters; Step 3: Calculate a comprehensive risk score based on the uterine contraction interval time series, predicted values ​​of physical sign parameters, and biochemical indicators; construct a personalized physiological baseline model for pregnant women, and output a personalized physiological baseline; Step 4: Implement a three-level early warning based on the uterine contraction interval time series, the predicted value of the physical sign parameter and the comprehensive risk score.

Images