Fetus safety monitoring method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of fetal monitoring, in particular to a fetal safety monitoring method and system based on multi-source data fusion. The method comprises the following steps: acquiring multi-source monitoring data; carrying out adaptive wavelet entropy noise reduction processing on the fetal heart rate signal, and extracting fetal heart rate features according to a processed result; quantifying uterine contraction energy based on the maternal uterine contraction pressure signal, and obtaining uterine contraction characteristics through the uterine contraction energy; an exponential decay weight algorithm of a uterine contraction time window is introduced, and blood oxygen features are extracted based on fetal blood oxygen signals and fetal movement data; obtaining motion degree characteristics according to the motion posture data of the pregnant woman; fusing the fetal heart rate feature, the uterine contraction feature, the blood oxygen feature and the motion degree feature to complete fetal safety monitoring. According to the method, the multi-source data is subjected to targeted special processing, then the automatic evaluation model is constructed to realize accurate and quantitative evaluation of the safety state of the fetus, and the problems that a single data source is easily interfered and depends on doctor experience in safety monitoring of the fetus are solved.

Description

Technical Field

[0001] This invention relates to the field of fetal monitoring technology, specifically to a fetal safety monitoring method and system based on multi-source data fusion. Background Technology

[0002] Fetal safety monitoring, as a core component of modern perinatal medicine to ensure maternal and infant health and safety, carries the hopes and aspirations of countless families for their newborns. It provides clinicians with valuable decision-making information, enabling effective interventions to be taken before the fetus suffers irreversible damage, thus building a solid defense for maternal and infant safety.

[0003] However, looking at current clinical fetal monitoring systems, mainstream methods still largely rely on single-dimensional monitoring data. These include fetal heart rate and its variability recorded by widely used fetal heart rate monitoring (such as NST, CST, and OCT), or growth indicators like biparietal diameter, femur length, and abdominal circumference, as well as information on amniotic fluid and placenta provided by regular ultrasound examinations. While these single data sources can provide valuable clues in specific areas, their limitations are becoming increasingly apparent in the complex and ever-changing intrauterine environment.

[0004] Taking fetal heart rate monitoring as an example, it mainly reflects the fetal intrauterine hypoxia and acid-base balance by continuously recording changes in fetal heart rate. However, this indicator is easily affected by a variety of internal and external factors: changes in maternal position, emotional fluctuations, the intensity of uterine contractions, and even eating and activity can cause temporary changes in fetal heart rate; the fetal sleep-wake cycle itself significantly affects the monitoring results. When the fetus is in a quiet sleep period, the baseline variability of the fetal heart rate decreases, which may lead to "false positive" results and increase unnecessary clinical intervention; and in some cases, even if the fetus has mild hypoxia, if it does not reach the level of stimulating the fetal heart rate to produce significant deceleration or tachycardia, the fetal heart rate monitoring may show "normal", thus resulting in "false negatives" and delaying the diagnosis.

[0005] These inherent limitations directly lead to the need for further improvement in the accuracy, sensitivity, specificity, timeliness, and continuity of existing fetal monitoring methods in terms of overall assessment. They are like viewing a leopard through a tube, failing to comprehensively and dynamically depict the true condition of the fetus in utero. When a single indicator is abnormal, clinicians often need to rely on experience or wait for confirmation from other test results. This not only increases diagnostic uncertainty but may also delay early identification, accurate diagnosis, and timely and effective treatment of fetal abnormalities at critical moments, thus posing a potential threat to the fetus's health and even life. Therefore, overcoming the limitations of single data points and constructing a multi-dimensional, intelligent, real-time dynamic fetal safety monitoring system has become a crucial issue urgently needing to be addressed in the field of perinatal medicine. Summary of the Invention

[0006] To address the shortcomings of existing methods and the needs of practical applications, and to solve the aforementioned problems, this invention provides a fetal safety monitoring method based on multi-source data fusion, comprising the following steps: acquiring multi-source monitoring data, including fetal heart rate signals, maternal uterine contraction pressure signals, fetal blood oxygenation signals, and maternal movement posture data; performing adaptive wavelet packet entropy denoising processing on the fetal heart rate signals, and extracting fetal heart rate features based on the processed results; quantifying uterine contraction energy based on the maternal uterine contraction pressure signals, and obtaining uterine contraction features through the uterine contraction energy; introducing an exponential decay weighting algorithm based on uterine contraction time windows, and extracting blood oxygenation features based on the fetal blood oxygenation signals and fetal movement data; obtaining movement intensity features based on the maternal movement posture data; and fusing the fetal heart rate features, uterine contraction features, blood oxygenation features, and movement intensity features to complete fetal safety monitoring.

[0007] This invention addresses the issues of susceptibility to interference from single data sources and reliance on physician experience in fetal safety monitoring. It integrates four types of multi-source data: fetal heart rate signals, maternal uterine contraction pressure signals, fetal blood oxygenation signals, and maternal movement posture data. By first performing targeted special processing on each type of data to extract key anti-interference features, then fusing the features in a hierarchical manner, and finally constructing an automated assessment model, it achieves accurate and quantitative assessment of fetal safety status.

[0008] Optionally, the adaptive wavelet packet entropy denoising process on the fetal heart rate signal, and the extraction of fetal heart rate features based on the processed result, includes the following steps: Based on the average frequency and peak factor of the fetal heart rate signal, the wavelet basis is adaptively determined; based on the wavelet basis, the entropy correction factor is calculated to obtain the wavelet dynamic threshold; through the wavelet dynamic threshold, the heart rate variability and artifact residue index are extracted; the present invention dynamically selects the wavelet basis by signal characteristics (frequency, impulsivity), which improves the separation degree between the decomposed signal and noise, and is beneficial for processing high fluctuation signals and strong artifact signals.

[0009] Optionally, the step of quantifying uterine contraction energy based on the maternal uterine contraction pressure signal and obtaining uterine contraction characteristics through the uterine contraction energy includes the following steps: Positional compensation correction is performed based on the body tilt angle, and uterine contraction energy is calculated according to the correction results. Using this uterine contraction energy, the position-corrected uterine contraction energy intensity ratio, the coefficient of variation of contraction intervals, and the positional uterine contraction synergy index are obtained. This invention reduces the error in assessing uterine contraction energy in the supine position through trunk tilt angle calculation and a cosine compensation factor.

[0010] Optionally, the exponential decay weighted algorithm that introduces a uterine contraction time window to extract blood oxygenation features based on the fetal blood oxygenation signal and fetal movement data includes the following steps: A fetal movement influencing factor was constructed, and an exponential decay weighting function was built based on this factor. The blood oxygenation signal within the contraction time window was then weighted and smoothed using this exponential decay weighting function. The lowest blood oxygenation value during contraction, the blood oxygenation recovery slope, and the fetal movement-blood oxygenation synergy index were calculated based on the processed blood oxygenation signal. By introducing the fetal movement influencing factor and correlation analysis, the signal-to-noise ratio of the blood oxygenation signal during contraction was improved, which is beneficial for more accurately extracting the characteristics of blood oxygenation changes during contraction.

[0011] Optionally, obtaining the movement intensity characteristics based on the pregnant woman's movement posture data includes the following steps: Based on the pregnant woman's movement posture data, the movement intensity is calculated using a movement intensity model. The mean movement intensity and the intensity of movement are then obtained. Combining these two factors, movement interference levels are categorized. Correction coefficients are constructed based on these interference levels, and a comprehensive movement interference index is obtained by combining these coefficients with the proportion of artifact interval durations. By dynamically adjusting the correction coefficients according to the movement interference levels, the correlation between the corrected features and the fetal safety status is further improved. Combining the interference level with the artifact proportion provides a more comprehensive quantitative basis for subsequent feature fusion, which is beneficial for improving the stability of the fused features.

[0012] Optionally, the process of fusing the fetal heart rate characteristics, uterine contraction characteristics, blood oxygenation characteristics, and movement intensity characteristics to complete fetal safety monitoring includes the following steps: By combining the fetal heart rate characteristics and the uterine contraction characteristics, a three-dimensional correlation feature of body position-heart rate-uterine contraction is obtained; by combining the blood oxygenation characteristics and the activity level characteristics, a blood oxygenation-activity correlation feature is obtained; a multi-dimensional feature attention mechanism safety monitoring model is constructed, and fetal safety monitoring is completed through the safety monitoring model based on the three-dimensional correlation feature of body position-heart rate-uterine contraction and the blood oxygenation-activity correlation feature. This transforms multi-source features into accurate assessment results, overcomes the limitations of single data, achieves automated and highly robust fetal safety monitoring, and reduces reliance on physician experience.

[0013] Optionally, the step of combining the fetal heart rate characteristics and the uterine contraction characteristics to obtain the three-dimensional correlation characteristics of body position-heart rate-uterine contraction includes the following steps: This invention utilizes a positional correction factor to correct the correlation between the positional-corrected uterine contraction energy intensity ratio and heart rate variability, and obtains a first fusion weight parameter based on the corrected correlation. A first position-heart rate-uterine contraction feature is obtained based on the first fusion weight parameter, the positional-corrected uterine contraction energy intensity ratio, and heart rate variability. A second position-heart rate-uterine contraction feature is obtained by combining heart rate variability and the position-uterine contraction synergy index. Finally, a three-dimensional correlation feature of position-heart rate-uterine contraction is obtained using the first and second position-heart rate-uterine contraction features. This invention addresses the deficiency of traditional two-dimensional fusion methods that ignore positional interference by constructing a three-dimensional correlation feature of position-heart rate-uterine contraction, thus improving the correlation between the fusion feature and fetal distress.

[0014] Optionally, the step of combining the blood oxygenation characteristics and the exercise intensity characteristics to obtain the blood oxygenation-exercise correlation characteristics includes the following steps: A dynamic compensation model is constructed using a motion sensitivity index. Blood oxygen urgency is categorized based on the lowest blood oxygen level during uterine contractions. Combining this urgency with the dynamic compensation model, blood oxygen-motion correlation characteristics are obtained. This invention constructs a synergistic compensation mechanism for fetal movement and motion interference, resolving the distortion of blood oxygen characteristics caused by their superposition, thus improving the accuracy of chronic hypoxia identification.

[0015] Optionally, the construction of the security monitoring model using a multi-dimensional feature attention mechanism includes the following steps: Dynamic attention weights are calculated using a normalized function; based on these dynamic attention weights, a dynamic weighted risk assessment model is constructed. This model utilizes an attention mechanism to enable it to possess human-like cognitive abilities, such as "dynamically focusing on high-value features," surpassing the fixed-weight model of traditional machine learning and more closely resembling the decision-making logic of clinicians.

[0016] Secondly, to efficiently execute the fetal safety monitoring method based on multi-source data fusion provided by this invention, this invention also provides a fetal safety monitoring system based on multi-source data fusion, comprising: a fetal heart rate signal module for extracting fetal heart rate signals; a maternal uterine contraction pressure signal module for extracting maternal uterine contraction pressure signals; a fetal blood oxygen signal module for extracting fetal blood oxygen signals; a maternal movement posture module for extracting maternal movement posture data; and a data processing module for extracting features from the fetal heart rate signal, the maternal uterine contraction pressure signal, the fetal blood oxygen signal, and the maternal movement posture data, and using the features to complete fetal safety monitoring. This fetal safety monitoring system based on multi-source data fusion is compact in structure and stable in performance, and can stably execute the fetal safety monitoring method based on multi-source data fusion provided by this invention, further improving the overall applicability and practical application capability of this invention. Attached Figure Description

[0017] Figure 1 A flowchart of a fetal safety monitoring method based on multi-source data fusion provided in an embodiment of the present invention; Figure 2 This is a framework diagram of a fetal safety monitoring system based on multi-source data fusion, provided for an embodiment of the present invention. Detailed Implementation

[0018] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0019] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0020] Please see Figure 1 To address the aforementioned problems, this invention provides a fetal safety monitoring method based on multi-source data fusion, such as... Figure 1 As shown, in one embodiment, the method includes the following steps: S1. Acquire multi-source monitoring data, including fetal heart rate signal, maternal uterine contraction pressure signal, fetal blood oxygen signal, and pregnant woman's movement posture data.

[0021] In this embodiment, an abdominal electrode is used to collect the raw fetal heart rate signal; a uterine contraction sensor is used to collect the raw maternal uterine contraction signal, and a triaxial accelerometer is used to simultaneously collect the pregnant woman's trunk tilt information (angle with the horizontal plane); near-infrared spectroscopy is used to collect the fetal blood oxygen saturation signal; and a triaxial accelerometer is used to collect the pregnant woman's motion data, including acceleration in the x, y, and z directions, to identify motion artifacts that may affect other signals.

[0022] S2. Perform adaptive wavelet packet entropy denoising on the fetal heart rate signal, and extract fetal heart rate features based on the processed results.

[0023] Fetal heart rate signals are easily affected by the pregnant woman's movement and breathing, resulting in motion artifacts and baseline drift, which distort the signal. First, the raw signal is preprocessed to remove high-frequency noise above 50Hz. Then, linear interpolation is used to fill in brief missing segments (less than 0.5 seconds in duration) to ensure signal continuity.

[0024] Further, the adaptive wavelet packet entropy denoising process performed on the fetal heart rate signal, and the extraction of fetal heart rate features based on the processed result, includes the following steps: S21. Based on the average frequency and peak factor of the fetal heart rate signal, adaptively determine the wavelet basis.

[0025] First, a candidate wavelet basis library is preset, including db4 (good smoothness), sym5 (good symmetry), and coif3 (fast decay). Then, an adaptive rule for selecting the wavelet basis is set. In this embodiment, the adaptive rule for the wavelet basis satisfies: When the average frequency is less than 1.2 Hz and the peak factor is less than 2.5, the db4 wavelet basis is selected. When the average frequency is greater than or equal to 1.2 Hz and the peak factor is less than 2.5, the sym5 wavelet basis is selected. When the peak factor is greater than or equal to 2.5, the coif3 wavelet basis is selected.

[0026] The gust factor reflects the signal's impulsiveness; motion artifacts can cause the gust factor to exceed 3. The gust factor satisfies the following formula: S22. Calculate the entropy correction factor based on the wavelet basis to obtain the wavelet dynamic threshold.

[0027] The number of decomposition layers is dynamically adjusted based on the average frequency. The higher the average frequency, the more decomposition layers are needed (4-6 layers) to ensure that high-frequency artifacts fall into independent frequency bands.

[0028] Furthermore, the frequency band entropy value is calculated to obtain the intra-layer average entropy. The entropy correction factor is then calculated using the intra-layer average entropy, satisfying the following conditions: in, Indicates the first Entropy correction factor for wavelet threshold. This represents the average entropy within the layer. Indicates gestational age. Indicates the wavelet basis fit. This represents the average value function.

[0029] Furthermore, a threshold is set individually for each frequency band to obtain the wavelet dynamic threshold, which satisfies the following: , This indicates a fixed threshold.

[0030] S23. Extract heart rate variability and artifact persistence index using the wavelet dynamic threshold.

[0031] Specifically, heart rate variability is extracted using the wavelet dynamic threshold, satisfying the following formula: in, Indicates the first The variability of the layer reconstructed signal, This represents the frequency band entropy value.

[0032] The artifact persistence index is used to quantitatively assess the noise reduction effect on fetal heart rate signals. The artifact persistence index is extracted and satisfies the following formula: in, This represents the noise-reduced fetal heart rate signal. This represents the original noisy signal. Represents time series sampling points, This represents the total number of sampling points. By counting the number of abnormal points where the signal difference before and after noise reduction exceeds 5 times / minute (bpm), the proportion of these abnormal points in the total sample is calculated and converted into a percentage. Clinical practice shows that when the artifact residue index is below 8%, the quality of the FHR signal after noise reduction can meet the needs of fetal safety monitoring and effectively reduce the risk of misjudgment caused by artifact interference.

[0033] S3. Quantify the contraction energy based on the maternal contraction pressure signal, and obtain the contraction characteristics through the contraction energy.

[0034] The raw signals of maternal uterine contractions collected by the uterine contraction pressure sensor include low-frequency interference such as maternal abdominal wall tension and respiration, resulting in baseline drift. Furthermore, the signal is significantly affected by the pregnant woman's position. When the woman is in a supine position, the sensor readings may be artificially high due to the pressure of the uterus on the abdominal wall, making it difficult to accurately determine the intensity and interval of uterine contractions.

[0035] The process of quantifying uterine contraction energy based on the maternal uterine contraction pressure signal and obtaining uterine contraction characteristics through the uterine contraction energy includes the following steps: S31. Perform positional compensation correction based on body tilt angle, and calculate uterine contraction energy based on the correction results.

[0036] First, the angle between the torso and the horizontal plane is calculated based on the acceleration data. Then, a cosine compensation factor is constructed based on the body tilt angle, satisfying the following formula: This represents the cosine compensation factor. Indicates the included angle. The position sensitivity coefficient is a parameter used to adjust the weight of the influence of body position on uterine contraction signals. Its core function is to balance the proportion of physical position factors and physiological uterine contraction signals. This coefficient is obtained by fitting clinical data, and in this embodiment, it is set to 0.7.

[0037] Then, a cosine compensation factor is used to perform positional compensation correction on the original signal. After positional correction, baseline adjustment is performed, and a 10-second sliding window is used to calculate the baseline to obtain the final corrected signal, which satisfies the following: Furthermore, a 30-second sliding window is used to amplify the peak energy of contractions through squaring operations, while simultaneously integrating duration information. The contraction energy function satisfies: Furthermore, the peak energy of uterine contractions and the time interval between adjacent peaks are extracted, and effective uterine contractions are screened using a dynamic threshold method, with the threshold adaptively adjusted according to the patient's position. In this embodiment, the dynamic threshold method refers to a peak energy of uterine contractions greater than 50 times the cosine compensation factor.

[0038] S32. Using the uterine contraction energy, obtain the position-corrected uterine contraction energy intensity ratio, the coefficient of variation of uterine contraction interval, and the position-based uterine contraction synergy index.

[0039] Specifically, the ratio of uterine contraction energy intensity corrected by body position satisfies the following formula: A sine term is introduced to compensate for the effect of decreased uterine contraction transmission when the patient is in the upright position.

[0040] The coefficient of variation of the contraction interval is used to quantify the dispersion of the contraction interval time, and satisfies the following formula: in, This represents the standard deviation of the time interval between uterine contractions after positional correction, reflecting the fluctuation range of the contraction interval. This represents the average interval between contractions after correction, reflecting the average cycle of contractions.

[0041] The Postural Coordination Index aims to reveal the correlation between changes in pregnant women's position and fluctuations in contraction intensity, satisfying the following formula: in, The rate of change in the pregnant woman's posture at time t is calculated by fusing data from an accelerometer and a gyroscope, reflecting the dynamic process of the pregnant woman's posture adjustment. The energy change rate of uterine contractions at time t is represented by the frequency domain analysis of the contraction pressure sensor signal, characterizing the dynamic change in contraction intensity. When the pregnant woman's position is relatively stable, uterine contraction activity should be mainly regulated by physiological mechanisms, and the coordination index should be at a low level. Clinical studies have shown that the threshold of this index should be less than 0.3 in a normal pregnancy. If it exceeds this range, it may indicate that abnormal activity of the pregnant woman is interfering with uterine contraction monitoring, or potential abnormal uterine contractions, requiring further comprehensive evaluation in conjunction with other monitoring indicators.

[0042] S4. An exponential decay weighting algorithm is introduced to incorporate the uterine contraction time window to extract blood oxygenation features based on the fetal blood oxygenation signal and fetal movement data.

[0043] Fetal blood oxygen saturation signals acquired using near-infrared spectroscopy show a characteristic decrease during uterine contractions due to changes in placental blood flow. However, these signals are easily affected by changes in fetal position and fetal movement, resulting in significant fluctuations. Fetal activity signals acquired using an abdominal accelerometer show characteristic acceleration peaks during fetal movement, which can be used to mark the timing of fetal movement.

[0044] The exponential decay weighted algorithm that introduces a uterine contraction time window extracts blood oxygenation features based on the fetal blood oxygenation signal and fetal movement data, including the following steps: S41. Construct a fetal movement influencing factor, and construct an exponential decay weight function based on the fetal movement influencing factor.

[0045] Specifically, the peak contraction time determined based on the above steps Defining the time window for labor contractions: This covers the critical period from 20 seconds before contractions to 40 seconds after contractions.

[0046] The fetal movement acceleration signal is processed and a fetal movement threshold is set to mark the fetal movement period. In this embodiment, the fetal movement threshold is set to 0.15 times the gravitational acceleration. Furthermore, the constructed fetal movement influence factor satisfies: Furthermore, an exponentially decaying weighting function is constructed based on the aforementioned fetal movement influencing factors, satisfying the following formula: This weighting function focuses on changes in blood oxygenation during the period of strongest uterine contractions, while also reducing the weight of signals during periods of fetal movement.

[0047] S42. The blood oxygen signal within the uterine contraction time window is weighted and smoothed using the exponential decay weighting function.

[0048] Weighted smoothing of blood oxygenation signals within the time window enhances the reliability of blood oxygenation signals during uterine contractions and non-fetal movement periods, while also satisfying the following requirements: S43. Calculate the lowest blood oxygen value during uterine contractions, the blood oxygen recovery slope, and the fetal movement-blood oxygen synergy index based on the processed blood oxygen signal.

[0049] Specifically, the number of fetal movements within the contraction time window is counted. Changes in blood oxygen levels within 10 seconds of fetal movement .

[0050] Furthermore, the lowest blood oxygen level during uterine contractions satisfies the following formula: The slope of blood oxygen recovery satisfies the following formula: The fetal movement-oxygenation synergy index satisfies the following formula: Fetal movement is generally considered to be a distracting factor. This method not only suppresses its interference but also uses it as an auxiliary indicator reflecting the fetus's reserve capacity to correct blood oxygenation characteristics, thus fully exploring the value of fetal movement data.

[0051] S5. Based on the pregnant woman's movement posture data, obtain the movement degree characteristics.

[0052] In this embodiment, obtaining the movement intensity characteristics based on the pregnant woman's movement posture data includes the following steps: S51. Based on the pregnant woman's movement posture data, the exercise intensity is calculated using an exercise intensity model. The mean exercise intensity and the intensity of exercise are obtained through the exercise intensity.

[0053] Specifically, the motion intensity model satisfies: Indicates exercise intensity. These are the accelerations in the x, y, and z directions, respectively.

[0054] The intensity of the exercise satisfies: This indicates the total monitoring duration.

[0055] S52. Based on the average exercise intensity and the intensity of the exercise, classify the level of exercise interference.

[0056] In this embodiment, the level of motion interference is determined based on the average motion intensity and the degree of motion severity. It is divided into three levels.

[0057] Specifically, if the average exercise intensity is less than 0.2 times the gravitational acceleration and the intensity of the exercise is less than 5%, the interference level is 1; If the average exercise intensity is greater than 0.2 times and less than 0.4 times the gravitational acceleration, and the intensity of the exercise is greater than 5% and less than 15%, the interference level is 2. If the average exercise intensity is greater than 0.4 times the gravitational acceleration and the intensity of the exercise is greater than 15%, the interference level is 3. S53. Based on the motion interference level, construct a correction coefficient, and combine the correction coefficient with the proportion of artifact interval duration to obtain a comprehensive motion interference index.

[0058] In the embodiment, the correction coefficient satisfy: Set exercise intensity threshold Mark the motion artifact interval Calculate the proportion of artifact interval duration .

[0059] The motion interference comprehensive index satisfies: By integrating interference levels and artifact proportions, the overall degree of interference of motion on features is quantified, providing a more comprehensive basis for interference quantification for subsequent feature fusion, which is conducive to improving the stability of fused features.

[0060] Furthermore, fetal heart rate and blood oxygenation characteristics affected by motion artifacts can be corrected using the following correction formula: The more severe the motion interference (the higher the level, the greater the proportion of artifacts), the greater the correction amplitude, to avoid under-correction or over-correction caused by fixed parameters under different interference intensities.

[0061] S6. By integrating the fetal heart rate characteristics, uterine contraction characteristics, blood oxygenation characteristics, and movement intensity characteristics, fetal safety monitoring is completed.

[0062] In this embodiment, the process of fusing the fetal heart rate characteristics, uterine contraction characteristics, blood oxygenation characteristics, and activity level characteristics to complete fetal safety monitoring includes the following steps: S61. Combining the fetal heart rate characteristics and the uterine contraction characteristics, obtain the three-dimensional correlation characteristics of body position-heart rate-uterine contraction.

[0063] First, the correlation between the positional correction factor and the ratio of uterine contraction energy intensity and heart rate variability is corrected using the positional correction factor, and the first fusion weight parameter is obtained based on the corrected correlation.

[0064] Based on the classical correlation coefficient, a weighted correlation coefficient model is constructed that satisfies: When the mother is in a supine position, the position significantly affects the transmission of uterine contraction signals, and the correlation weight between fetal heart rate variability and uterine contraction signals is enhanced through a coefficient adjustment mechanism; while in an upright position, the weight automatically decreases to reflect the actual physiological coupling characteristics.

[0065] Furthermore, the first fusion weight parameter is obtained based on the corrected correlation, satisfying: Secondly, based on the first fusion weight parameters, the position-corrected uterine contraction energy intensity ratio, and heart rate variability, the first position-heart rate-uterine contraction characteristics are obtained.

[0066] Specifically, the first body position-heart rate-uterine contraction characteristics satisfy: This indicates the variability of average heart rate.

[0067] Then, by combining heart rate variability and position-contraction synergy index, the second position-heart rate-contraction characteristics were obtained.

[0068] Specifically, the second body position-heart rate-uterine contraction characteristics satisfy: Finally, the three-dimensional correlation features of body position-heart rate-uterine contraction are obtained through the first body position-heart rate-uterine contraction features and the second body position-heart rate-uterine contraction features.

[0069] Depend on and Together they form a two-dimensional feature vector This vector, by fusing time-series features (fetal heart rate variability), energy features (positive energy ratio), and spatial features (position-contraction synergy), achieves three-dimensional correlation modeling of position, fetal heart rate, and contraction signals, providing richer and more discriminative feature representations for subsequent intelligent analysis of fetal status.

[0070] S62. Combining the blood oxygenation characteristics and the exercise intensity characteristics, obtain the blood oxygenation-exercise correlation characteristics.

[0071] First, a dynamic compensation model is constructed using the motion sensitivity index.

[0072] Introducing a motion sensitivity index to satisfy: This represents the standardized value for gestational age, ranging from 0 to 1. This represents the standardized body mass index value, ranging from 0 to 1, reflecting the pregnant woman's sensitivity to exercise disturbances.

[0073] Furthermore, the dynamic compensation model satisfies: Secondly, the oxygen urgency level is classified according to the lowest blood oxygen value during uterine contractions. Combining the oxygen urgency level with the dynamic compensation model, the blood oxygen-motor correlation characteristics are obtained.

[0074] The clinical importance of minimum blood oxygen saturation and recovery rate changes dynamically with fetal status (e.g., minimum blood oxygen saturation is more critical in acute hypoxia, while recovery rate better reflects reserve capacity in chronic hypoxia). Fixed weights are difficult to apply. Therefore, classifying blood oxygen saturation urgency based on the minimum blood oxygen saturation during contractions should meet the following criteria: Furthermore, by combining the aforementioned blood oxygen urgency level and the aforementioned dynamic compensation model, the blood oxygen-exercise correlation characteristics are obtained, satisfying: This indicates the weights of blood oxygen data and exercise data. Indicates average exercise intensity. Indicates the rate of blood oxygen recovery. This represents the fetal movement pressure signal. The fetal movement pressure signal is first standardized (divided by the maximum value to convert to a 0-1 range) to eliminate the influence of absolute dimensions, retaining only relative intensity information to ensure compatibility with blood oxygenation characteristics (percentage) and recovery rate (% / s). Through non-linear weighting, the inherent relationship between blood oxygenation and movement is preserved while quantifying fetal movement interference. S63. Construct a safety monitoring model with a multi-dimensional feature attention mechanism. Based on the three-dimensional correlation features of body position-heart rate-uterine contraction and the correlation features of blood oxygen-motor activity, the safety monitoring model is used to complete fetal safety monitoring.

[0075] In this embodiment, constructing a security monitoring model with a multi-dimensional feature attention mechanism includes the following steps: First, a normalized function is used to calculate the dynamic attention weights.

[0076] The dynamic attention weights are calculated using the softmax normalization function, satisfying the following: in, This represents the weight of the i-th attention head. Let the feature importance index of the i-th feature be denoted by: in, The Gini index represents the i-th feature. The clinical expert assessment index for the i-th feature is determined by the obstetric team's scoring of the feature's clinical relevance to fetal safety. For example, the clinical expert assessment index value for the uterine contraction energy index is usually higher than 0.8.

[0077] Then, based on the dynamic attention weights, a dynamic weighted risk assessment model is constructed.

[0078] Specifically, the dynamic weighted risk assessment model satisfies: in, This represents the regression output of the gradient boosting decision tree model for each feature. It automatically learns complex nonlinear relationships between features through the gradient boosting tree structure. This represents the bias term, used to calibrate the overall prediction bias of the model. Over-attention weights are used to weight and aggregate the output of the gradient boosting decision tree model, enabling the model to dynamically focus on highly reliable features based on data quality, significantly reducing the risk of misjudgments caused by artifacts and motion interference.

[0079] Furthermore, using the three-dimensional correlation features of body position-heart rate-uterine contraction, blood oxygen-motor correlation features, artifact residual index, and motion interference comprehensive index as inputs, a dynamic weighted risk assessment model is used to obtain fetal safety assessment values. Thresholds are set according to gestational age: <0.65 in early pregnancy is considered high risk, <0.7 in mid-pregnancy, and <0.75 in late pregnancy, to achieve accurate judgment.

[0080] Please see Figure 2 In this embodiment, to efficiently execute the fetal safety monitoring method based on multi-source data fusion provided by the present invention, the present invention also provides a fetal safety monitoring system based on multi-source data fusion, comprising: a fetal heart rate signal module for extracting fetal heart rate signals; a maternal uterine contraction pressure signal module for extracting maternal uterine contraction pressure signals; a fetal blood oxygen signal module for extracting fetal blood oxygen signals; a maternal movement posture module for extracting maternal movement posture data; and a data processing module for extracting features from the fetal heart rate signal, the maternal uterine contraction pressure signal, the fetal blood oxygen signal, and the maternal movement posture data, and using the features to complete fetal safety monitoring. The fetal safety monitoring system based on multi-source data fusion of the present invention has a compact structure and stable performance, and can stably execute the fetal safety monitoring method based on multi-source data fusion of the present invention, further improving the overall applicability and practical application capability of the present invention.

[0081] The present invention also provides a fetal safety monitoring device based on multi-source data fusion, comprising: an input device, an output device, a processor, and a memory, wherein the input device, the output device, the processor, and the memory are interconnected, and the memory contains program instructions for the steps of the fetal safety monitoring method based on multi-source data fusion.

[0082] In this embodiment, the processor may be a central processing unit, but it can also be other general-purpose processors, digital signal processors, application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (OPGs), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. Input devices can be used to acquire data. Output devices can be used to output the results obtained by storing program instructions contained in a computer program in the memory provided by this invention. The memory may include read-only memory and random access memory (RAM), and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory (RAM).

[0083] In one possible implementation, the memory may include a stored program area and a stored data area. The stored program area may store the operating system and applications required for at least one function; the stored data area may store data created during use. Furthermore, the memory may include read-only memory and random access memory, and provides instructions and data to the processor. The memory stores the operating system and operating instructions, executable modules, or data structures, or subsets thereof, or extended sets thereof. The operating instructions may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and handling hardware-based tasks.

[0084] The embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described fetal safety monitoring method based on multi-source data fusion.

[0085] The storage medium can include various media that can store program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0086] In summary, this invention addresses the issues of susceptibility to interference from single data sources and reliance on physician experience in fetal safety monitoring. It integrates four types of multi-source data: fetal heart rate signals, maternal uterine contraction pressure signals, fetal blood oxygenation signals, and maternal movement posture data. By first performing targeted special processing on each type of data to extract key anti-interference features, then fusing the features in a hierarchical manner, and finally constructing an automated assessment model, it achieves accurate and quantitative assessment of fetal safety status.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. 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 or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the present invention.

Claims

1. A method for fetal safety monitoring based on multi-source data fusion, characterized in that, The following steps are involved: Acquire multi-source monitoring data, including fetal heart rate signal, maternal uterine contraction pressure signal, fetal blood oxygen signal, and pregnant woman's movement posture data; The fetal heart rate signal is subjected to adaptive wavelet packet entropy denoising processing, and fetal heart rate features are extracted based on the processing results. The contraction energy is quantified based on the maternal contraction pressure signal, and the contraction characteristics are obtained through the contraction energy. An exponential decay weighted algorithm for uterine contraction time windows is introduced to extract blood oxygenation features based on the fetal blood oxygenation signal and fetal movement data; Based on the pregnant woman's movement posture data, the movement degree characteristics are obtained; By integrating the fetal heart rate characteristics, uterine contraction characteristics, blood oxygenation characteristics, and activity level characteristics, fetal safety monitoring is completed.

2. The fetal safety monitoring method based on multi-source data fusion according to claim 1, characterized in that, The adaptive wavelet packet entropy denoising process for the fetal heart rate signal, and the extraction of fetal heart rate features based on the processed result, includes the following steps: Based on the average frequency and peak factor of the fetal heart rate signal, the wavelet basis is adaptively determined; The entropy correction factor is calculated based on the wavelet basis to obtain the wavelet dynamic threshold; Heart rate variability and artifact persistence index are extracted using the wavelet dynamic threshold.

3. The fetal safety monitoring method based on multi-source data fusion according to claim 1, characterized in that, The process of quantifying uterine contraction energy based on the maternal uterine contraction pressure signal and obtaining uterine contraction characteristics through the uterine contraction energy includes the following steps: Postural compensation and correction are performed based on the body tilt angle, and uterine contraction energy is calculated based on the correction results. Using the contraction energy, the position-corrected contraction energy intensity ratio, the contraction interval variation coefficient, and the position-contraction synergy index are obtained.

4. The fetal safety monitoring method based on multi-source data fusion according to claim 1, characterized in that, The exponential decay weighted algorithm that introduces a uterine contraction time window extracts blood oxygenation features based on the fetal blood oxygenation signal and fetal movement data, including the following steps: Construct a fetal movement influencing factor, and build an exponentially decaying weight function based on the fetal movement influencing factor; The blood oxygenation signal within the uterine contraction time window is weighted and smoothed using the exponentially decaying weight function. The lowest blood oxygen level during uterine contractions, the slope of blood oxygen recovery, and the fetal movement-blood oxygen synergy index were calculated based on the processed blood oxygenation signal.

5. The fetal safety monitoring method based on multi-source data fusion according to claim 1, characterized in that, The process of obtaining movement intensity characteristics based on the pregnant woman's movement posture data includes the following steps: Based on the pregnant woman's exercise posture data, the exercise intensity is calculated using an exercise intensity model. The mean exercise intensity and the degree of exercise intensity are obtained from the exercise intensity. Based on the average exercise intensity and the degree of exercise severity, the level of exercise interference is classified; Based on the motion interference level, a correction coefficient is constructed, and by combining the correction coefficient and the proportion of artifact interval duration, a comprehensive motion interference index is obtained.

6. The fetal safety monitoring method based on multi-source data fusion according to claim 1, characterized in that, The process of integrating fetal heart rate characteristics, uterine contraction characteristics, blood oxygenation characteristics, and activity level characteristics to complete fetal safety monitoring includes the following steps: By combining the fetal heart rate characteristics and the uterine contraction characteristics, a three-dimensional correlation feature of body position-heart rate-uterine contraction is obtained; By combining the blood oxygenation characteristics and the exercise intensity characteristics, blood oxygenation-exercise correlation characteristics are obtained; A multi-dimensional feature attention mechanism safety monitoring model is constructed. Based on the three-dimensional correlation features of body position-heart rate-uterine contraction and the correlation features of blood oxygen-motor activity, the safety monitoring model is used to complete fetal safety monitoring.

7. The fetal safety monitoring method based on multi-source data fusion according to claim 6, characterized in that, The process of combining the fetal heart rate characteristics and the uterine contraction characteristics to obtain the three-dimensional correlation characteristics of body position-heart rate-uterine contraction includes the following steps: The correlation between the postural correction factor and the postural correction uterine contraction energy intensity ratio and heart rate variability was corrected, and the first fusion weight parameter was obtained based on the corrected correlation. Based on the first fusion weight parameters, the position-corrected uterine contraction energy intensity ratio, and heart rate variability, the first position-heart rate-uterine contraction characteristics are obtained. By combining heart rate variability and the position-contraction synergy index, the second position-heart rate-contraction characteristics were obtained. The three-dimensional correlation features of body position-heart rate-uterine contraction are obtained by using the first body position-heart rate-uterine contraction features and the second body position-heart rate-uterine contraction features.

8. The fetal safety monitoring method based on multi-source data fusion according to claim 6, characterized in that, The process of combining the blood oxygenation characteristics and the exercise intensity characteristics to obtain blood oxygenation-exercise correlation characteristics includes the following steps: A dynamic compensation model is constructed using the motion sensitivity index; Blood oxygen urgency is determined based on the lowest blood oxygen level during uterine contractions. By combining the blood oxygen urgency with the dynamic compensation model, blood oxygen-motor correlation characteristics are obtained.

9. The fetal safety monitoring method based on multi-source data fusion according to claim 6, characterized in that, The security monitoring model constructed using a multi-dimensional feature attention mechanism includes the following steps: Dynamic attention weights are calculated using a normalized function; Based on the dynamic attention weights, a dynamic weighted risk assessment model is constructed.

10. A fetal safety monitoring system based on multi-source data fusion, characterized in that, include: A fetal heart rate signal module, which is used to extract the fetal heart rate signal; A maternal uterine contraction pressure signal module, which is used to extract maternal uterine contraction pressure signals; A fetal blood oxygenation signal module, wherein the fetal blood oxygenation signal module is used to extract fetal blood oxygenation signals; A pregnant woman's movement posture module, which is used to extract pregnant woman's movement posture data; The data processing module is used to extract features from the fetal heart rate signal, the maternal uterine contraction pressure signal, the fetal blood oxygen signal and the pregnant woman's movement posture data, and use the features to complete fetal safety monitoring.

Images