A predictive and interventional decision-making method and related equipment for the risk of stillbirth associated with umbilical cord torsion.

Abstract

This invention discloses a method and related equipment for predicting and intervening in the risk of stillbirth related to umbilical cord torsion. The method acquires multimodal clinical data, including the pregnant woman's complaints and prenatal ultrasound examination information, and standardizes the data to extract core predictive factors such as changes in fetal movement, signs of umbilical cord root torsion, and fetal growth restriction, as well as auxiliary predictive factors such as abnormal amniotic fluid volume and abnormal placental-umbilical cord insertion, constructing a feature vector that the model can process. Based on a pre-trained risk prediction model, the risk of stillbirth related to umbilical cord torsion is quantitatively assessed, a comprehensive risk score is output, and risk levels are classified. Based on the risk level, a pre-set clinical intervention strategy is automatically matched to form standardized treatment recommendations, which are then output to the clinical terminal. This method overcomes the limitations of traditional methods that rely on low sensitivity of single ultrasound diagnosis, achieving early risk identification and tiered management.

Description

Technical Field

[0001] This specification relates to the field of health management, and more specifically, this application relates to a method and related equipment for predicting and intervening in the risk of stillbirth associated with umbilical cord torsion. Background Technology

[0002] Umbilical cord torsion (UCT) is a serious pregnancy complication that can restrict or even interrupt umbilical cord blood flow and is a major cause of stillbirth. Because the umbilical cord is the only channel through which the fetus receives oxygen and nutrients, once umbilical cord torsion occurs, it often progresses rapidly and has serious clinical consequences.

[0003] Currently, prenatal assessment of umbilical cord torsion in clinical practice mainly relies on prenatal ultrasound. However, due to limitations such as fetal position, the concealment of umbilical cord movement, and image resolution, conventional prenatal ultrasound has limited diagnostic capabilities for umbilical cord torsion. Based on over ten years of clinical data research at our hospital, the inventors found that the diagnostic sensitivity of conventional prenatal ultrasound for umbilical cord torsion is only about 7.4%. The vast majority of cases are difficult to identify in time during the prenatal stage, leading to missed intervention opportunities and potentially serious adverse outcomes such as intrauterine fetal death.

[0004] The existing technology has the following main shortcomings: First, clinical decision-making relies too heavily on ultrasound examination, but ultrasound has a high rate of missing diagnoses of umbilical cord torsion, resulting in a significant blind spot in risk identification; second, there is a lack of standardized predictive tools that can comprehensively assess the risk of stillbirth related to umbilical cord torsion by integrating multidimensional clinical information such as the pregnant woman's complaints, fetal growth status, and specific ultrasound signs; third, clinical intervention is highly dependent on the doctor's personal experience and lacks a unified decision-making path based on large-sample evidence-based medicine, leading to inconsistencies in management and intervention strategies.

[0005] Therefore, there is an urgent need for a technical solution that can overcome the limitations of traditional ultrasound diagnosis, integrate multimodal clinical data, enable early prediction of umbilical cord torsion-related prenatal stillbirth risk, and provide standardized decision support for clinical practice. Summary of the Invention

[0006] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0007] Firstly, this application proposes a method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion, including: Acquire multimodal clinical data of pregnant women and standardize the multimodal clinical data, wherein the multimodal clinical data includes at least the pregnant woman's chief complaint information and prenatal ultrasound examination information; Core and auxiliary predictive factors for predicting the risk of stillbirth related to umbilical cord torsion were extracted from standardized multimodal clinical data. The aforementioned core predictive factors and auxiliary predictive factors are encoded into feature vectors that the model can process. The aforementioned feature vectors are input into a pre-trained risk prediction model, and a comprehensive risk score for stillbirths related to umbilical cord torsion is calculated based on the aforementioned risk prediction model. The risk level of the above comprehensive risk score is classified according to the preset risk threshold; Based on the aforementioned risk levels, the system automatically matches the corresponding intervention decision-making plan from the pre-set clinical intervention strategy library and outputs risk assessment results and intervention recommendations to guide clinical treatment.

[0008] In one feasible implementation, the core predictor and auxiliary predictor factors extracted from standardized multimodal clinical data for predicting the risk of stillbirth related to umbilical cord torsion include: Information on changes in fetal movement was extracted from the above-mentioned information on the pregnant women's chief complaints, and this information on changes in fetal movement was used as one of the core predictive factors. The torsion sign of the umbilical cord at the insertion point of the fetal abdominal wall was extracted from the above prenatal ultrasound examination information, and the torsion sign of the umbilical cord root was used as one of the core predictive factors. Fetal growth and development assessment results are extracted from the above prenatal ultrasound examination information. When the estimated fetal weight is lower than the preset percentile threshold for the same gestational age, the above fetal growth restriction information is used as one of the above core predictive factors. Meanwhile, the amniotic fluid volume assessment results and placental umbilical cord insertion morphology assessment results were extracted from the above prenatal ultrasound examination information and used as the above auxiliary predictive factors.

[0009] In one feasible implementation, encoding the core predictor and the auxiliary predictor into a model-processable feature vector includes: The aforementioned core predictive factors and auxiliary predictive factors are discretized or quantified respectively to form a feature representation with a unified dimension. Based on the preset feature coding rules, the above information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental umbilical cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables. The mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model.

[0010] In one feasible implementation, encoding the core predictor and the auxiliary predictor into a model-processable feature vector includes: The aforementioned core predictive factors and auxiliary predictive factors are discretized or quantified respectively to form a feature representation with a unified dimension. Based on the preset feature coding rules, the above information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental umbilical cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables. The mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model.

[0011] In one feasible implementation, before encoding the core predictor and the auxiliary predictor into a model-processable feature vector, the method further includes: Based on the degree of influence of different predictors on the risk of stillbirth related to umbilical cord torsion, different weight levels are set for the core predictors and the auxiliary predictors. Among them, the weight level of the core predictors is higher than that of the auxiliary predictors, the weight level of the information on changes in fetal movement in pregnant women is higher than that of signs of umbilical cord torsion, and the weight level of signs of umbilical cord torsion is higher than that of information on fetal growth restriction. In the above feature vector construction process, the weight levels are encoded as feature parameters into the feature vector.

[0012] In one feasible implementation, the aforementioned feature vector is input into a pre-trained risk prediction model, and a comprehensive risk score for umbilical cord torsion-related stillbirth is calculated based on the risk prediction model, including: The model structure for constructing the aforementioned risk prediction model includes at least: a feature input layer for receiving the feature vector composed of the aforementioned core predictive factors and the aforementioned auxiliary predictive factors; a feature weighting layer for weighting each feature component in the aforementioned feature vector according to predetermined model coefficients; a risk calculation layer for performing linear combination operations on the weighted feature results and generating intermediate risk values; and a probability mapping layer for performing nonlinear mapping on the aforementioned intermediate risk values ​​and outputting the predicted probability value of umbilical cord torsion-related stillbirths. The aforementioned feature vectors are input into the aforementioned feature input layer, and in the aforementioned feature weighting layer, the aforementioned core predictive factors are assigned a higher weight coefficient than the aforementioned auxiliary predictive factors based on the weight levels of different predictive factors. In the aforementioned risk calculation layer, the weighted feature components are summed to obtain the intermediate risk value that characterizes the intensity of stillbirth risk. In the above probability mapping layer, the intermediate risk value is transformed based on a preset probability function to obtain the predicted probability value of the occurrence of stillbirth related to umbilical cord torsion. Based on the preset probability and scoring mapping rules, the predicted probability values ​​are converted into the comprehensive risk score.

[0013] In one feasible implementation, based on the aforementioned risk level, the system automatically matches a corresponding intervention decision-making plan from a pre-set clinical intervention strategy library and outputs risk assessment results and intervention recommendations to guide clinical management, including: When the comprehensive risk score is higher than the first preset threshold, it is determined to be a high-risk level, and an intervention decision-making plan including immediate hospitalization, continuous fetal heart rate monitoring and preparation for emergency termination of pregnancy is automatically matched. When the comprehensive risk score is between the first and second preset thresholds, it is determined to be of medium risk level, and intervention decision-making plans including shortening the re-examination cycle, increasing the frequency of prenatal monitoring, and implementing further functional assessment are automatically matched. When the comprehensive risk score is lower than the second preset threshold, it is determined to be a low-risk level, and an intervention decision-making plan including routine prenatal check-up management and enhanced guidance for pregnant women's self-monitoring is automatically matched. The aforementioned risk levels, comprehensive risk scores, and intervention decision-making plans will be output to clinical terminals in a visual or structured format.

[0014] Secondly, this invention also proposes a prediction and intervention decision-making system for the risk of stillbirth related to umbilical cord torsion, comprising: An acquisition unit is used to acquire multimodal clinical data of pregnant women and to standardize the multimodal clinical data, wherein the multimodal clinical data includes at least the pregnant woman's chief complaint information and prenatal ultrasound examination information. The extraction unit is used to extract core and auxiliary predictive factors for predicting the risk of stillbirth related to umbilical cord torsion from standardized multimodal clinical data. An encoding unit is used to encode the core predictor and the auxiliary predictor into a feature vector that the model can process; The input unit is used to input the feature vector into a pre-trained risk prediction model and calculate a comprehensive risk score for stillbirths related to umbilical cord torsion based on the risk prediction model. A grading unit is used to classify the comprehensive risk score into risk levels based on a preset risk threshold. The matching unit is used to automatically match the corresponding intervention decision plan from the preset clinical intervention strategy library according to the risk level, and output the risk assessment results and intervention suggestions to guide clinical treatment.

[0015] Thirdly, the present invention also proposes an electronic device comprising: a memory and a processor, characterized in that the processor is used to execute a computer program stored in the memory to implement the steps of the prediction and intervention decision-making method for umbilical cord torsion-related stillbirth risk as described in any of the first aspects.

[0016] Fourthly, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the prediction and intervention decision-making method for umbilical cord torsion-related stillbirth risk as described in any of the first aspects.

[0017] In summary, the prediction and intervention decision-making method for umbilical cord torsion-related stillbirth risk provided in this application has achieved significant improvements in risk identification methods, assessment accuracy, and clinical decision support capabilities, demonstrating significant technical effects and clinical value. Firstly, addressing the problem of excessive reliance on prenatal ultrasound in existing technologies, where ultrasound has extremely low sensitivity for diagnosing umbilical cord torsion, this method is no longer limited to single imaging evidence. Instead, it combines multimodal clinical data such as the pregnant woman's complaints, key ultrasound signs, and fetal growth status for joint analysis. In particular, it incorporates fetal movement changes, which are highly sensitive to fetal hypoxia, into the core predictive factor system. This allows for effective early warning of potential high-risk conditions even before imaging abnormalities are clearly identified, enabling proactive identification of umbilical cord torsion-related stillbirth risk and mitigating the clinical safety loopholes caused by the high rate of missed diagnoses in traditional ultrasound. Secondly, this method standardizes and vectorizes multi-source clinical data, transforming clinical information, originally presented as textual descriptions or experiential judgments, into a unified, fixed-dimensional, and directly computable model input. This not only avoids the impact of differences in recording methods among different doctors and systems on risk assessment results but also ensures good repeatability and consistency in the risk assessment process, laying a reliable data foundation for subsequent model inference and clinical application. Thirdly, by introducing a pre-trained risk prediction model, this method can quantitatively model the risk contribution of different predictive factors based on large historical sample data and output continuous comprehensive risk scores. Compared to traditional subjective judgments relying on personal experience, this represents a shift from "experience-driven" to "data-driven and evidence-driven," significantly reducing the uncertainty and arbitrariness of clinical decision-making. Furthermore, by setting risk thresholds, this method transforms continuous risk scores into clear risk levels, allowing complex model outputs to be presented to clinicians as intuitive and actionable hierarchical signals. This reduces reliance on understanding the model's internal parameters and improves practical operability. This method automatically matches risk levels with a pre-defined clinical intervention strategy library, achieving closed-loop management from risk assessment to intervention recommendation output. This allows high-risk pregnant women to be identified promptly and quickly enter intensive monitoring or emergency intervention procedures, medium-risk pregnant women to receive more intensive follow-up and confirmatory assessments, and low-risk pregnant women to receive standardized management focused on health education and self-monitoring. This ensures safety while avoiding unnecessary over-intervention. In summary, this method, through multimodal data fusion, risk quantification modeling, and standardized decision output, effectively overcomes the problems of identification lag, strong subjectivity in assessment, and inconsistent intervention pathways in existing technologies for assessing stillbirth risk related to umbilical cord torsion. It significantly improves the scientific rigor, standardization, and clinical feasibility of prenatal stillbirth risk management.

[0018] Other advantages, objectives and features of this application will be apparent in part from the description which follows, and in part from what those skilled in the art will understand through study and practice of this application. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A schematic flowchart illustrating a method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion, provided in an embodiment of this application. Figure 2 A structural schematic diagram of a decision-making system for predicting and intervening in the risk of stillbirth related to umbilical cord torsion, provided in an embodiment of this application; Figure 3 This is a schematic diagram of an electronic device structure provided in an embodiment of this application. Detailed Implementation

[0020] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0021] Please see Figure 1 This is a flowchart illustrating a method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion, as provided in an embodiment of this application. Specifically, it may include: S110. Obtain multimodal clinical data of pregnant women and standardize the multimodal clinical data, wherein the multimodal clinical data includes at least the pregnant woman's chief complaint information and prenatal ultrasound examination information. S120. Extract core and auxiliary predictive factors for predicting the risk of stillbirth related to umbilical cord torsion from standardized multimodal clinical data. S130. Encode the above core predictive factors and the above auxiliary predictive factors into feature vectors that the model can process. S140. Input the above feature vector into the pre-trained risk prediction model, and calculate the comprehensive risk score of stillbirth related to umbilical cord torsion based on the above risk prediction model. S150. Classify the risk level of the above comprehensive risk score according to the preset risk threshold. S160. Based on the above risk levels, automatically match the corresponding intervention decision plan from the preset clinical intervention strategy library, and output the risk assessment results and intervention suggestions to guide clinical treatment.

[0022] For example, in step S110, the pregnant woman's multimodal clinical data is first acquired and standardized. The multimodal clinical data includes at least the pregnant woman's chief complaint information and prenatal ultrasound examination information. The purpose of this step is to unify clinical information from different sources and with different forms of expression into a calculable and comparable structured data format. For example, chief complaints such as "decreased / absent fetal movement" are converted into standard options or quantitative levels, and indicators such as "umbilical cord root torsion," "amniotic fluid volume," "placental cord insertion morphology," and "estimated fetal weight / gestational age" in the ultrasound report are converted into standardized fields. The consistency of missing values, outliers, and timestamps is checked, thereby providing a consistent input for subsequent factor extraction and model inference.

[0023] In step S120, core predictive factors and auxiliary predictive factors for predicting the risk of stillbirth related to umbilical cord torsion are extracted from the standardized multimodal clinical data. The core predictive factors preferentially reflect key features that are highly indicative of UCT-related stillbirths, typically including decreased or absent fetal movement reported by the pregnant woman, signs of torsion at the umbilical cord root (fetal insertion site in the abdominal wall) indicated by ultrasound, and information on fetal growth restriction obtained by ultrasound or weight assessment. Auxiliary predictive factors are used to supplement the risk profile and improve the robustness of the model, typically including abnormal amniotic fluid volume (such as polyhydramnios) and abnormal placental cord insertion. By combining "highly sensitive subjective signals + specific imaging indications + fetal growth status", a more reliable comprehensive indication of stillbirth risk can still be formed even when conventional ultrasound is insufficiently sensitive to UCT.

[0024] In step S130, the aforementioned core predictive factors and auxiliary predictive factors are encoded into feature vectors that the model can process. This process is not just a simple "yes / no" assignment, but can also be discretized, hierarchically coded, and subject to rule constraints based on clinical definitions. For example, fetal movement change information is encoded as a binary or graded variable, umbilical cord root torsion signs are encoded as "present / absent" or "mild / severe" categories, fetal growth restriction information is formed into a standard judgment position based on the estimated weight percentile threshold, and amniotic fluid volume and placental umbilical cord insertion morphology are mapped to consistent category variables. At the same time, feature vectors are formed by concatenating them in a fixed field order to ensure that different cases are completely consistent in input dimensions, meanings, and units, and to avoid model inference bias caused by differences in description.

[0025] In step S140, the aforementioned feature vectors are input into a pre-trained risk prediction model, and a comprehensive risk score for stillbirths related to umbilical cord torsion is calculated based on this model. In implementation, the risk prediction model can employ a highly interpretable structured model such as logistic regression, which internally performs inference through feature input, feature weighting, risk calculation, and probability mapping. The model first receives the feature vectors and weights each feature according to the coefficients obtained during training, thereby forming an intermediate risk value. Then, it obtains the predicted probability of stillbirth through nonlinear mapping and outputs a comprehensive risk score of 0-100 according to a preset probability-score mapping rule. Because the core predictive factors have a stronger risk contribution in the training samples, they naturally exhibit higher weights at the model coefficient level. This results in a significant increase in the score when key features such as "reduced / absent fetal movement," "signs of umbilical cord torsion," and "fetal growth restriction" occur simultaneously, thus achieving sensitive capture of high-risk conditions.

[0026] In step S150, the comprehensive risk score is classified into risk levels based on a preset risk threshold. This step transforms the continuous score into a stratified management signal that can be directly implemented in clinical practice. For example, the score can be mapped to low-risk, medium-risk, and high-risk levels. The threshold is set by previous clinical validation data or expert consensus to ensure that the stratification is both safe and operable. Through the stratification mechanism, clinicians can quickly grasp the risk intensity without interpreting complex model parameters.

[0027] In step S160, based on the aforementioned risk levels, the corresponding intervention decision-making plan is automatically matched from the preset clinical intervention strategy library, and risk assessment results and intervention recommendations are output to guide clinical treatment. The key to this step is to close the loop from "prediction" to "intervention". The strategy library pre-stores treatment paths corresponding to different levels in the form of standardized rules. For example, high risk triggers a higher-level warning and recommends immediate hospitalization, continuous fetal heart rate monitoring, initiating a high-level obstetric consultation and assessing the feasibility of emergency termination of pregnancy. Medium risk recommends shortening the interval between follow-up examinations, increasing the frequency of monitoring, and improving further functional assessment. Low risk outputs recommendations such as routine prenatal check-ups and strengthening education on fetal movement self-monitoring. At the same time, the output end presents the score, level, interpretation of key risk factors, and recommended treatment points on the doctor's workstation, mobile terminal, or web interface, thereby reducing the subjective fluctuation of decision-making and ensuring that high-risk pregnant women are identified in a timely manner and receive consistent standardized management.

[0028] In summary, the prediction and intervention decision-making method for umbilical cord torsion-related stillbirth risk provided in this application has achieved significant improvements in risk identification methods, assessment accuracy, and clinical decision support capabilities, demonstrating significant technical effects and clinical value. Firstly, addressing the problem of excessive reliance on prenatal ultrasound in existing technologies, where ultrasound has extremely low sensitivity for diagnosing umbilical cord torsion, this method is no longer limited to single imaging evidence. Instead, it combines multimodal clinical data such as the pregnant woman's complaints, key ultrasound signs, and fetal growth status for joint analysis. In particular, it incorporates fetal movement changes, which are highly sensitive to fetal hypoxia, into the core predictive factor system. This allows for effective early warning of potential high-risk conditions even before imaging abnormalities are clearly identified, enabling proactive identification of umbilical cord torsion-related stillbirth risk and mitigating the clinical safety loopholes caused by the high rate of missed diagnoses in traditional ultrasound. Secondly, this method standardizes and vectorizes multi-source clinical data, transforming clinical information, originally presented as textual descriptions or experiential judgments, into a unified, fixed-dimensional, and directly computable model input. This not only avoids the impact of differences in recording methods among different doctors and systems on risk assessment results but also ensures good repeatability and consistency in the risk assessment process, laying a reliable data foundation for subsequent model inference and clinical application. Thirdly, by introducing a pre-trained risk prediction model, this method can quantitatively model the risk contribution of different predictive factors based on large historical sample data and output continuous comprehensive risk scores. Compared to traditional subjective judgments relying on personal experience, this represents a shift from "experience-driven" to "data-driven and evidence-driven," significantly reducing the uncertainty and arbitrariness of clinical decision-making. Furthermore, by setting risk thresholds, this method transforms continuous risk scores into clear risk levels, allowing complex model outputs to be presented to clinicians as intuitive and actionable hierarchical signals. This reduces reliance on understanding the model's internal parameters and improves practical operability. This method automatically matches risk levels with a pre-defined clinical intervention strategy library, achieving closed-loop management from risk assessment to intervention recommendation output. This allows high-risk pregnant women to be identified promptly and quickly enter intensive monitoring or emergency intervention procedures, medium-risk pregnant women to receive more intensive follow-up and confirmatory assessments, and low-risk pregnant women to receive standardized management focused on health education and self-monitoring. This ensures safety while avoiding unnecessary over-intervention. In summary, this method, through multimodal data fusion, risk quantification modeling, and standardized decision output, effectively overcomes the problems of identification lag, strong subjectivity in assessment, and inconsistent intervention pathways in existing technologies for assessing stillbirth risk related to umbilical cord torsion. It significantly improves the scientific rigor, standardization, and clinical feasibility of prenatal stillbirth risk management.

[0029] In one feasible implementation, the core predictor and auxiliary predictor factors extracted from standardized multimodal clinical data for predicting the risk of stillbirth related to umbilical cord torsion include: Information on changes in fetal movement was extracted from the above-mentioned information on the pregnant women's chief complaints, and this information on changes in fetal movement was used as one of the core predictive factors. The torsion sign of the umbilical cord at the insertion point of the fetal abdominal wall was extracted from the above prenatal ultrasound examination information, and the torsion sign of the umbilical cord root was used as one of the core predictive factors. Fetal growth and development assessment results are extracted from the above prenatal ultrasound examination information. When the estimated fetal weight is lower than the preset percentile threshold for the same gestational age, the above fetal growth restriction information is used as one of the above core predictive factors. Meanwhile, the amniotic fluid volume assessment results and placental umbilical cord insertion morphology assessment results were extracted from the above prenatal ultrasound examination information and used as the above auxiliary predictive factors.

[0030] For example, the system first extracts information on changes in fetal movement from the pregnant woman's chief complaint, and uses this information as one of the core predictive factors. This is because changes in fetal movement directly reflect the dynamic changes in the fetus's intrauterine state and are among the most sensitive early signals of acute risks such as fetal hypoxia and circulatory restriction. In implementation, the information on changes in fetal movement is not limited to a simple "present / absent" state, but can further include structured fields such as the duration of decreased fetal movement, the magnitude of the decrease, or the subjective severity reported by the pregnant woman. These are then standardized into calculable categories or levels according to unified rules, ensuring that chief complaints from different doctors and with different recording methods can still be consistently interpreted and used for risk assessment. Simultaneously, the system extracts signs of umbilical cord torsion at the insertion point in the fetal abdominal wall from the prenatal ultrasound examination information, and uses this umbilical cord root torsion as one of the core predictive factors. The emphasis on the root region of "fetal abdominal wall insertion point" is because tight torsion at this location is more likely to lead to restricted umbilical cord blood flow, impaired venous return, or umbilical cord compression, thus significantly increasing the risk of acute intrauterine hypoxia and even stillbirth.

[0031] In practice, the extraction of the above-mentioned signs of umbilical cord root torsion can be completed through the structuring of ultrasound report keywords, import of image annotation results, or selective entry of doctors' workstations. Furthermore, the morphological characteristics of the torsion (such as whether the torsion is tight, whether it is accompanied by local blood flow abnormalities, etc.) can be recorded to enhance the expression accuracy. However, at the factor level, it remains a unified field to facilitate stable model calling.

[0032] Furthermore, the system extracts fetal growth and development assessment results from the aforementioned prenatal ultrasound examination information. When the estimated fetal weight is lower than the preset percentile threshold for the same gestational age, the aforementioned fetal growth restriction information is used as one of the core predictive factors. This is because fetal growth restriction usually indicates a risk of chronic placental insufficiency, limited nutrient supply, or persistent hypoxia. In the context of pathological mechanisms affecting blood flow, such as umbilical cord torsion, the aforementioned fetal growth restriction information can serve as important evidence of "long-term risk accumulation," thus forming a complementary relationship of "acute signal + chronic background" with the aforementioned information on changes in fetal movement. In implementation, the preset percentile threshold can be selected from clinically commonly used thresholds such as the 10th percentile for the same gestational age. It is also required that the fetal weight estimation calculation method, gestational age determination method, and reference growth curve source be consistent within the system to avoid misjudgments due to inconsistent standards.

[0033] In addition to the core predictive factors mentioned above, the system also extracts amniotic fluid volume assessment results and placental-umbilical cord insertion morphology assessment results from the prenatal ultrasound examination information, which are used as auxiliary predictive factors. The amniotic fluid volume assessment results reflect indirect changes in fetal urinary circulation and placental function, especially in cases of polyhydramnios, which may indicate abnormal load on the fetal-placental system or other high-risk factors. As an auxiliary signal, this improves the model's coverage of complex cases. The placental-umbilical cord insertion morphology assessment results characterize the structural basis of the umbilical cord-placental connection. For example, abnormal insertion such as marginal insertion or velamentous attachment may lead to changes in the umbilical cord's stress pattern, increased local blood flow vulnerability, or increased susceptibility to torsion / compression, thus providing a "structural vulnerability" explanation for umbilical cord torsion-related risks.

[0034] By identifying the aforementioned information on changes in fetal movement, signs of umbilical cord root torsion, and information on fetal growth restriction as core predictive factors, and the aforementioned assessment results of amniotic fluid volume and the assessment results of placental umbilical cord insertion morphology as auxiliary predictive factors, this implementation method achieves hierarchical extraction of risk information and organization of the chain of evidence: core predictive factors are used to drive major changes in risk scores and ensure sensitive capture of high-risk states, while auxiliary predictive factors are used to supplement individual differences and pathological background and enhance model robustness, thereby laying a unified, interpretable, and clinically applicable data foundation for subsequent feature vector encoding, risk prediction model inference, and intervention decision matching corresponding to risk levels.

[0035] In one feasible implementation, encoding the core predictor and the auxiliary predictor into a model-processable feature vector includes: The aforementioned core predictive factors and auxiliary predictive factors are discretized or quantified respectively to form a feature representation with a unified dimension. Based on the preset feature coding rules, the above information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental umbilical cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables. The mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model.

[0036] For example, encoding the core predictive factors and auxiliary predictive factors into feature vectors that the model can process essentially transforms clinical semantic information into a "computable, alignable, and reusable" structured input, enabling stable input into the risk prediction model and consistent output of a comprehensive risk score. Since the core and auxiliary predictive factors include both "yes / no" indicators and potentially range values, grade descriptions, or textual conclusions, this embodiment achieves consistent representation of cross-source data through a combination of discretization, numerical representation, and rule mapping.

[0037] First, the core predictive factors and auxiliary predictive factors mentioned above are discretized or quantified to form a unified feature representation. Specifically, discretization is used to unify common unstructured or semi-structured clinical expressions into finite categories. For example, complaints such as "decreased fetal movement, absent fetal movement, normal fetal movement" are grouped into discrete levels, "presence / absence of umbilical cord root torsion" is grouped into binary categories, and "normal placental cord insertion morphology, marginal insertion, velamentous attachment" is grouped into multi-value categories. Quantification is used to convert quantifiable indicators into directly computable numerical features. For example, the percentile of fetal weight estimation relative to the reference curve for the same gestational age is converted into a continuous value, or the amniotic fluid volume assessment results are expressed as standard quantitative indicators such as the amniotic fluid index (AFI) or maximum amniotic fluid pool depth. Through the above processing, the recording habits of different doctors, the reporting formats of different examination equipment, and the field calibers of different systems can be unified before entering the model, thereby reducing the impact of input noise on model inference.

[0038] Based on preset feature coding rules, the above information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables.

[0039] In practical implementation, the aforementioned preset feature encoding rules can be defined using a "field-value-encoding" triplet format, ensuring that each clinical meaning corresponds to a unique encoded expression. For example, the aforementioned information on changes in fetal movement in pregnant women can be mapped to a binary variable (0 indicates no / normal, 1 indicates yes / decreased or disappeared), or to a multi-valued variable (0 indicates normal, 1 indicates decreased, 2 indicates disappeared) to enhance the precision of expression. The aforementioned sign of umbilical cord root torsion can be mapped to a binary variable (0 indicates not seen, 1 indicates seen), or it can be mapped to a multi-valued variable based on the degree of tightness. The aforementioned information on fetal growth restriction is usually triggered by "fetal weight estimate being lower than the preset percentile threshold for the same gestational age," and can be mapped to a binary variable (0 indicates the threshold is not met, 1 indicates the threshold is met), while the percentile can be retained simultaneously as a supplementary numerical feature. The aforementioned amniotic fluid volume assessment result can be mapped to a binary variable (0 indicates normal, 1 indicates abnormal, such as excessive), or to a multi-valued variable (0 normal, 1 low, 2 high) to cover more clinical scenarios. The above assessment results of placental umbilical cord insertion morphology can be mapped to multi-valued variables (0 normal, 1 marginal insertion, 2 velamentous attachment, 3 other abnormalities) to reflect the risk background brought about by structural differences. Through the above-mentioned regularized mapping, it is ensured that similar information from different cases is expressed in the same coding system, and it is also convenient for subsequent interpretation and tracing of the model's contribution.

[0040] Finally, the mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model. The key to this step lies in the "fixed order" and "uniform dimension," meaning that the dimension and field position of the feature vector remain unchanged regardless of whether certain examination results or fields are missing in a case. For missing items, default values ​​or missing marker values ​​can be filled according to a preset missing value handling strategy, thereby avoiding model inference failure or output bias due to inconsistent dimensions. Furthermore, the fixed order is usually strictly consistent with the feature order during model training, for example, it can be concatenated in the order of "chief complaint-related features—torsion sign features—growth restriction features—amniotic fluid volume features—insertion morphology features," so that the model can accurately correspond to the coefficient weights of each feature during calculation, ensuring the repeatability and verifiability of the risk score.

[0041] Through the above coding process, this embodiment achieves a standardized bridge from clinical semantics to model input: on the one hand, discretization / numerization and rule mapping ensure the consistency of multimodal data in expression; on the other hand, fixed order combination ensures the stability and reproducibility of model inference, thereby providing a reliable data foundation for subsequently inputting the above feature vectors into the above risk prediction model, calculating the comprehensive risk score, and further performing risk classification and intervention decision matching.

[0042] In one feasible implementation, encoding the core predictor and the auxiliary predictor into a model-processable feature vector includes: The aforementioned core predictive factors and auxiliary predictive factors are discretized or quantified respectively to form a feature representation with a unified dimension. Based on the preset feature coding rules, the above information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental umbilical cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables. The mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model.

[0043] In one feasible implementation, before encoding the core predictor and the auxiliary predictor into a model-processable feature vector, the method further includes: Based on the degree of influence of different predictors on the risk of stillbirth related to umbilical cord torsion, different weight levels are set for the core predictors and the auxiliary predictors. Among them, the weight level of the core predictors is higher than that of the auxiliary predictors, the weight level of the information on changes in fetal movement in pregnant women is higher than that of signs of umbilical cord torsion, and the weight level of signs of umbilical cord torsion is higher than that of information on fetal growth restriction. In the above feature vector construction process, the weight levels are encoded as feature parameters into the feature vector.

[0044] For example, the aforementioned core predictive factors and auxiliary predictive factors are discretized or quantified to form a unified feature representation. Discretization is used to compress "textual conclusions or category descriptions" into a finite set of values. For example, the aforementioned information on changes in fetal movement in pregnant women is discretized into levels such as "normal / decreased / disappeared," the aforementioned signs of umbilical cord root torsion are discretized into "absent / present" or "mild / severe," and the aforementioned assessment results of placental umbilical cord insertion morphology are discretized into "normal / marginal insertion / velamentous attachment / other abnormalities." Quantification is used to map quantifiable indicators into numerical features. For example, fetal weight percentiles and amniotic fluid index are represented as continuous values, or they are further thresholded into binary indicators to improve rule consistency. Through the above processing, similar information from different examination sources and different report formats has a unified data type and caliber before entering the coding stage.

[0045] Based on preset feature coding rules, the aforementioned information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental-umbilical cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables. For example, the preset feature coding rules can be defined in the form of "field-value-encoding," ensuring that each clinical meaning corresponds to a unique coded expression. For instance, the aforementioned information on changes in fetal movement can be mapped to... When fetal movement is normal When fetal movement decreases When fetal movement disappears The above signs of umbilical cord root torsion are mapped to Before seeing When it is visible (Or expanded to multiple values ​​according to severity); the above fetal growth restriction information is mapped as follows: When the estimated fetal weight is lower than the preset percentile threshold for the same gestational age ,otherwise The above amniotic fluid volume assessment results are mapped to Normally When there is too much amniotic fluid (Or expanded to multiple values ​​of less / normal / more); the above placental umbilical cord insertion morphology assessment results are mapped to For example, normal Marginal insertion sail-like attachment Other abnormalities Based on this, the mapped feature variables constitute the basic feature set. Its dimensions and meanings remained consistent across all cases.

[0046] Then, the mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model. This fixed order ensures that the feature positions are perfectly aligned during model inference and training. When individual fields are missing, default values ​​or missing marker values ​​can be filled according to preset missing data handling rules, thereby ensuring that the feature vector dimension remains unchanged and preventing model unavailability or output drift due to fluctuations in input dimension.

[0047] In one feasible implementation, before encoding the core predictor and auxiliary predictor into a feature vector that the model can process, the method further includes setting different weight levels for the core predictor and auxiliary predictor based on the degree of influence of different predictor factors on the risk of stillbirth related to umbilical cord torsion. These weight levels are then encoded as feature parameters into the feature vector during feature vector construction. The purpose of this design is to make the "prior structure of influence strength" explicit: on the one hand, the weight level of the core predictor is higher than that of the auxiliary predictor; on the other hand, within the core predictor, the weight level of information on changes in fetal movement is higher than that of signs of umbilical cord torsion, and the weight level of signs of umbilical cord torsion is higher than that of information on fetal growth restriction, thereby forming a hierarchical relationship consistent with clinical risk contribution.

[0048] For example, a weighting parameter can be set for each predictor. And by the weight level mapping function Generate numerical weights ,For example: in, Indicates the first The weight levels of the predictor factors, the This can be a monotonically increasing mapping, such that higher levels correspond to greater numerical weights. Furthermore, the weight levels can be directly encoded as weight vectors. and with the basic feature set Jointly construct extended feature vector ,For example in, The above feature vector is used as input to the risk prediction model. The first half is the value feature of the predictor, and the second half is the weight level feature of the corresponding predictor. This allows the model to use two types of information simultaneously when learning or reasoning: whether an event has occurred / to what extent and the importance of the event in risk assessment.

[0049] In another, more compact and computationally efficient implementation, the weight levels can also be used to perform weighted fusion of features to form weighted features. ,For example: in, This is a feature scale normalization or piecewise mapping function used to compress binary or multi-valued features to a uniform dimensional interval, for example, to preserve binary features as... Multi-valued features are normalized according to the highest level. In this case, the feature vector used as input to the aforementioned risk prediction model can be written as: This integrates weight level information into the features themselves, reducing vector dimensionality and enhancing the model's sensitivity to key factors.

[0050] The meanings of the parameters in the above formulas can be explained as follows: This represents the set of basic features mapped from the above information on changes in fetal movement in pregnant women, the above signs of umbilical cord root torsion, the above information on fetal growth restriction, the above results of amniotic fluid volume assessment, and the above results of placental and umbilical cord insertion morphology assessment. Indicates the first The encoded values ​​of each feature variable; Indicates the first The weighting levels of each predictor factor; This represents a monotonic mapping function that generates numerical weights from weight levels. Indicates the relationship with the first The numerical weights corresponding to each predictor factor; Represents the weight vector; This represents the extended feature vector obtained by jointly encoding the feature variables and weight variables; This represents a normalization or piecewise mapping function used to unify different characteristic dimensions; This represents the weighted feature after merging the weights; This represents the feature vector used as input to the model after weight fusion.

[0051] Through the aforementioned weighting and coding mechanisms, this embodiment can not only achieve unified characterization of multimodal clinical information, but also embed the priority of key risk signals into the feature construction process in a structured manner, so that subsequent model inference has more clinically logical response characteristics when facing different combination patterns, thereby improving the consistency, interpretability and clinical usability of risk prediction.

[0052] In one feasible implementation, the aforementioned feature vector is input into a pre-trained risk prediction model, and a comprehensive risk score for umbilical cord torsion-related stillbirth is calculated based on the risk prediction model, including: The model structure for constructing the aforementioned risk prediction model includes at least: a feature input layer for receiving the feature vector composed of the aforementioned core predictive factors and the aforementioned auxiliary predictive factors; a feature weighting layer for weighting each feature component in the aforementioned feature vector according to predetermined model coefficients; a risk calculation layer for performing linear combination operations on the weighted feature results and generating intermediate risk values; and a probability mapping layer for performing nonlinear mapping on the aforementioned intermediate risk values ​​and outputting the predicted probability value of umbilical cord torsion-related stillbirths. The aforementioned feature vectors are input into the aforementioned feature input layer, and in the aforementioned feature weighting layer, the aforementioned core predictive factors are assigned a higher weight coefficient than the aforementioned auxiliary predictive factors based on the weight levels of different predictive factors. In the aforementioned risk calculation layer, the weighted feature components are summed to obtain the intermediate risk value that characterizes the intensity of stillbirth risk. In the above probability mapping layer, the intermediate risk value is transformed based on a preset probability function to obtain the predicted probability value of the occurrence of stillbirth related to umbilical cord torsion. Based on the preset probability and scoring mapping rules, the predicted probability values ​​are converted into the comprehensive risk score.

[0053] For example, this embodiment constructs a hierarchical model structure, breaking down the model reasoning process into five consecutive steps: feature access, feature weighting, risk calculation, probability mapping, and score conversion. This makes the model output both interpretable and easy to deploy and reproduce stably in clinical information systems.

[0054] First, the model structure of the aforementioned risk prediction model is constructed, which includes at least a feature input layer, a feature weighting layer, a risk calculation layer, and a probability mapping layer. The feature input layer receives the feature vector composed of the core predictive factors and the auxiliary predictive factors, ensuring that the input dimension is consistent with that of the training phase. The feature weighting layer weights each feature component in the feature vector according to pre-determined model coefficients, thereby explicitly expressing the contribution of different features to the stillbirth risk. The risk calculation layer performs a linear combination operation on the weighted feature results and generates the intermediate risk value, enabling the model to generate a traceable quantification of risk intensity. The probability mapping layer performs a non-linear mapping on the intermediate risk value, outputting the predicted probability of stillbirth related to umbilical cord torsion, thus converting the linear risk intensity into a more clinically understandable "probability of occurrence."

[0055] For example, the above feature vector can be represented as: in, Indicates the first Each feature component can be encoded from the aforementioned information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental-umbilical cord insertion morphology assessment results. This represents the dimension of the aforementioned feature vector. To reflect the higher importance constraint of the core predictor compared to the auxiliary predictor, this embodiment may also introduce a weight level vector: And generate weight coefficients using a weight level mapping function: in, Indicates the first The weight levels of the predictor factors, the The mapping function is monotonically increasing, so that the higher the weight level, the corresponding... The larger the value, the higher the contribution of the core predictive factors in the weighting stage.

[0056] Subsequently, the aforementioned feature vector is input into the aforementioned feature input layer, and in the aforementioned feature weighting layer, the core predictive factor is assigned a higher weight coefficient than the aforementioned auxiliary predictive factor based on the weight levels of different predictive factors. For example, the aforementioned feature weighting layer can output a weighted feature vector: in, This represents the feature components modulated by the weighting coefficients. This represents the weighting coefficient derived from the weighting levels. Through this mechanism, even if different cases show abnormalities in certain auxiliary factors, the model can still preferentially respond to key risk signals such as changes in fetal movement, signs of root torsion, and growth restriction, thus better aligning with clinical cognitive pathways.

[0057] Next, in the aforementioned risk calculation layer, the weighted feature components are summed using a weighted summation operation to obtain the intermediate risk value characterizing the intensity of the stillbirth risk. For example, a linear combination form can be used: in, This represents the aforementioned intermediate risk value. Indicates the bias term. Indicates the first The model coefficients corresponding to each feature component are pre-trained and used to characterize the marginal contribution of each predictor factor to the stillbirth risk. This expression is interpretable: when a key feature appears and its corresponding... When it is large, This will increase significantly, thereby triggering a higher probability of risk and a higher score.

[0058] Then, in the aforementioned probability mapping layer, the intermediate risk value is transformed based on a preset probability function to obtain the predicted probability value of umbilical cord torsion-related stillbirth. For example, the probability function can be a sigmoid function to map the real number domain to... : in, This represents the predicted probability value of stillbirth related to umbilical cord torsion mentioned above. This represents the Sigmoid probability mapping function. This is a natural constant. Through this nonlinear mapping, the model can represent the risk intensity. Convert to intuitive probability This facilitates subsequent threshold grading and clinical interpretation.

[0059] Finally, based on the preset probability and scoring mapping rules, the predicted probability values ​​are converted into the comprehensive risk score, making the model output more consistent with the "score-based presentation and hierarchical management" usage habits in clinical workflows. For example, linear mapping or piecewise mapping rules can be used: in, The above comprehensive risk score indicates that... This represents a truncation function used to ensure that the score falls between 0 and 100; in another implementation, piecewise mapping can also be used to enhance the resolution of high-risk segments, for example... This makes the score more sensitive to changes in the high-risk probability range, which is more conducive to triggering clinical early warning and upgrading of intervention strategies.

[0060] The meanings of the parameters in the above formulas are as follows: This represents the feature vector used as input to the model; Indicates the first Each feature component; Represents a weight level vector; Indicates the first The weighting levels of each predictor factor; This represents a function that maps weight levels to numerical weights; This represents the corresponding weighting coefficient; This represents the weighted eigenvector after weight modulation; Indicates the bias term; Represents the model coefficients; Indicates the intermediate risk value; This represents the predicted probability value; Represents a probability mapping function; This represents the overall risk score; This represents the scoring range constraint function.

[0061] This embodiment can stably transform multimodal risk information into quantitative outputs that can be used for subsequent risk level classification and intervention decision matching while maintaining a clear and interpretable model structure, providing clinical practice with traceable, executable, and standardized UCT-related stillbirth risk assessment basis.

[0062] In one feasible implementation, based on the aforementioned risk level, the system automatically matches a corresponding intervention decision-making plan from a pre-set clinical intervention strategy library and outputs risk assessment results and intervention recommendations to guide clinical management, including: When the comprehensive risk score is higher than the first preset threshold, it is determined to be a high-risk level, and an intervention decision-making plan including immediate hospitalization, continuous fetal heart rate monitoring and preparation for emergency termination of pregnancy is automatically matched. When the comprehensive risk score is between the first and second preset thresholds, it is determined to be of medium risk level, and intervention decision-making plans including shortening the re-examination cycle, increasing the frequency of prenatal monitoring, and implementing further functional assessment are automatically matched. When the comprehensive risk score is lower than the second preset threshold, it is determined to be a low-risk level, and an intervention decision-making plan including routine prenatal check-up management and enhanced guidance for pregnant women's self-monitoring is automatically matched. The aforementioned risk levels, comprehensive risk scores, and intervention decision-making plans will be output to clinical terminals in a visual or structured format.

[0063] For example, the automatic matching of corresponding intervention decision-making schemes from the preset clinical intervention strategy library based on the aforementioned risk level, and the output of risk assessment results and intervention suggestions to guide clinical treatment, is a key closed-loop link in transforming "model quantitative output" into "executable clinical pathways". Its core lies in achieving risk stratification management through preset thresholds and mapping the stratification results to standardized intervention strategies in a rule-based manner, thereby reducing inconsistencies in treatment caused by differences in experience among different doctors and ensuring that high-risk pregnant women can receive timely, uniform and traceable intervention measures within a limited time window.

[0064] Specifically, the system first uses the comprehensive risk score as input, calls risk grading rules to segment the comprehensive risk score, and generates the risk level accordingly. This grading rule forms a clear three-segment division logic through a first preset threshold and a second preset threshold, ensuring a definite and reproducible correspondence between the score and the level, thereby guaranteeing consistent grading results under different times, terminals, and operator conditions. Simultaneously, the clinical intervention strategy library pre-stores standardized treatment strategy entries corresponding one-to-one with different risk levels. Each entry includes treatment objectives, execution time limits, monitoring intensity, re-examination frequency, escalation trigger conditions, and necessary reminder information, enabling the system to directly retrieve and output actionable clinical recommendations after completing the risk level determination, without relying on doctors' ad-hoc judgments or improvisation.

[0065] When the comprehensive risk score exceeds the first preset threshold, the system identifies it as a high-risk level and automatically matches intervention decision-making plans including immediate hospitalization, continuous fetal heart rate monitoring, and preparation for emergency termination of pregnancy. At this high-risk level, the system's strategy design emphasizes "time sensitivity" and "rapid escalation." This is because the pregnant woman has already exhibited multiple strong risk warning signals or a significantly increased model-predicted probability, and continued outpatient follow-up may miss the optimal intervention window. Therefore, the system prioritizes action recommendations with mandatory and high priority, such as recommending immediate hospitalization to ensure continuous monitoring, recommending continuous fetal heart rate monitoring to dynamically capture signs of fetal distress, and recommending the simultaneous initiation of preparations for emergency termination of pregnancy. This allows the medical team to complete crucial preparations such as preoperative assessment and coordination of anesthesia and surgical resources before the risk escalates further. For example, the clinical intervention strategy library can also include extended prompts such as "high-level physician consultation," "operating room alert," and "linkage of blood products and neonatal resuscitation resources" at this level, making the intervention decision-making plan more procedural and implementable.

[0066] When the comprehensive risk score falls between the first and second preset thresholds, the system identifies it as a medium-risk level and automatically matches intervention decision-making plans including shortening the follow-up examination cycle, increasing the frequency of prenatal monitoring, and implementing further functional assessments. At this medium-risk level, the system's strategy design emphasizes "strengthening monitoring and rapid verification," aiming to quickly confirm whether the risk has progressed and promptly escalate the treatment to a high-risk path by shortening the follow-up examination interval and increasing the monitoring density. Specifically, shortening the follow-up examination cycle can be expressed as a recommendation for follow-up examinations within a preset time limit; increasing the frequency of prenatal monitoring can be expressed as a recommendation for more frequent fetal heart rate monitoring or ultrasound follow-up; and further functional assessments are used to supplement the comprehensive judgment of the fetal intrauterine status, such as confirmatory assessments through biophysical scoring, umbilical artery Doppler blood flow monitoring, or other fetal functional assessment methods. Through this medium-risk intervention path, the system can not only reduce the risk of missed cases but also avoid excessive intervention for all suspected cases, thus balancing safety and the rational use of medical resources.

[0067] When the comprehensive risk score is lower than the second preset threshold, the system determines it to be at a low-risk level and automatically matches an intervention decision plan that includes routine prenatal check-up management and enhanced guidance for pregnant women's self-monitoring. At this low-risk level, the system's strategy design emphasizes a "health education and self-monitoring closed loop." This is because such pregnant women currently have a low risk, and excessive examinations may increase medical burden and psychological stress. Therefore, the system focuses on routine prenatal check-ups while simultaneously providing guidance on fetal movement self-monitoring. For example, it reminds pregnant women to observe and record fetal movements according to established methods and clarifies the triggering conditions and pathways for seeking medical attention when there is a significant decrease or disappearance of fetal movement or other abnormal symptoms. This transforms low-risk management from "passive waiting" to a closed-loop model of "active self-monitoring + timely feedback." For example, the clinical intervention strategy library can include educational points, precautions, and follow-up reminder templates at this level to improve compliance and feasibility.

[0068] After matching the risk level determination with the intervention decision plan, the system outputs the risk level, the comprehensive risk score, and the intervention decision plan to the clinical terminal in a visual or structured format. The visual format can be displayed on the doctor's workstation or mobile terminal interface using color-coded levels, risk bars, or warning boxes, along with a list of key actions and execution deadlines for the intervention decision plan on the same interface. The structured format can be output to the hospital information system or electronic medical record system with resolvable fields, such as a structured record containing fields for scoring, level, recommended treatment, suggested follow-up time, and escalation trigger conditions, facilitating record keeping, tracking, and quality control. Through this method, this embodiment achieves an automated closed-loop output from risk scoring to risk grading to intervention plans, thereby improving the standardization and clinical operability of umbilical cord torsion-related stillbirth risk management.

[0069] Secondly, this invention also proposes a prediction and intervention decision-making system for the risk of stillbirth related to umbilical cord torsion, such as... Figure 2 As shown, it includes: The acquisition unit 21 is used to acquire multimodal clinical data of pregnant women and to standardize the multimodal clinical data, wherein the multimodal clinical data includes at least the pregnant woman's chief complaint information and prenatal ultrasound examination information. Extraction unit 22 is used to extract core predictors and auxiliary predictors for predicting the risk of stillbirth related to umbilical cord torsion from standardized multimodal clinical data; Encoding unit 23 is used to encode the core predictor and the auxiliary predictor into a feature vector that the model can process; Input unit 24 is used to input the feature vector into a pre-trained risk prediction model and calculate a comprehensive risk score for stillbirths related to umbilical cord torsion based on the risk prediction model; The grading unit 25 is used to classify the risk level of the comprehensive risk score according to a preset risk threshold. The matching unit 26 is used to automatically match the corresponding intervention decision plan from the preset clinical intervention strategy library according to the risk level, and output the risk assessment results and intervention suggestions to guide clinical treatment.

[0070] In one feasible implementation, a decision-making system for predicting and intervening in the risk of stillbirth associated with umbilical cord torsion can also perform any step of the method proposed in the first aspect.

[0071] Thirdly, the present invention also proposes an electronic device 300, such as... Figure 3 As shown, it includes a memory 310, a processor 320, and a computer program 311 stored on the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of the prediction and intervention decision-making method for the risk of stillbirth related to umbilical cord torsion as described in any of the first aspects.

[0072] Fourthly, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the prediction and intervention decision-making method for umbilical cord torsion-related stillbirth risk as described in any one of the first aspects.

[0073] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0074] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0076] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0077] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0078] This application also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device performs the voice-based identity recognition process in the corresponding embodiment. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0079] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0080] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0081] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0082] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0083] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0084] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion, characterized in that, include: Acquire multimodal clinical data of pregnant women and standardize the multimodal clinical data, wherein the multimodal clinical data includes at least the pregnant woman's chief complaint information and prenatal ultrasound examination information; Core and auxiliary predictive factors for predicting the risk of stillbirth related to umbilical cord torsion were extracted from standardized multimodal clinical data. The core predictor and the auxiliary predictor are encoded into feature vectors that the model can process. The feature vector is input into a pre-trained risk prediction model, and a comprehensive risk score for stillbirths related to umbilical cord torsion is calculated based on the risk prediction model. The comprehensive risk score is classified into risk levels based on a preset risk threshold; Based on the risk level, the system automatically matches the corresponding intervention decision-making plan from the preset clinical intervention strategy library and outputs risk assessment results and intervention recommendations to guide clinical treatment.

2. The method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion as described in claim 1, characterized in that, The core predictive factors and auxiliary predictive factors extracted from standardized multimodal clinical data for predicting the risk of stillbirth related to umbilical cord torsion include: Information on changes in fetal movement is extracted from the pregnant woman's chief complaint information, and this information on changes in fetal movement is used as one of the core predictive factors. The torsion sign of the umbilical cord at the insertion point of the fetal abdominal wall is extracted from the prenatal ultrasound examination information, and the torsion sign of the umbilical cord root is used as one of the core predictive factors. Fetal growth and development assessment results are extracted from the prenatal ultrasound examination information. When the estimated fetal weight is lower than the preset percentile threshold for the same gestational age, the fetal growth restriction information is used as one of the core predictive factors. Meanwhile, the amniotic fluid volume assessment results and placental umbilical cord insertion morphology assessment results are extracted from the prenatal ultrasound examination information and used as auxiliary predictive factors, respectively.

3. The method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion as described in claim 1, characterized in that, The step of encoding the core predictor and the auxiliary predictor into a feature vector that the model can process includes: The core predictive factor and the auxiliary predictive factor are discretized or quantified respectively to form a feature representation with a unified dimension. Based on preset feature coding rules, the information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental umbilical cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables. The mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model.

4. The method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion as described in claim 1, characterized in that, The step of encoding the core predictor and the auxiliary predictor into a feature vector that the model can process includes: The core predictive factor and the auxiliary predictive factor are discretized or quantified respectively to form a feature representation with a unified dimension. Based on preset feature coding rules, the information on changes in fetal movement, signs of umbilical cord root torsion, information on fetal growth restriction, amniotic fluid volume assessment results, and placental umbilical cord insertion morphology assessment results are mapped to corresponding binary or multi-valued feature variables. The mapped feature variables are combined in a fixed order to generate the feature vector used as input to the risk prediction model.

5. The method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion according to claim 1, characterized in that, Before encoding the core predictor and the auxiliary predictor into a model-processable feature vector, the method further includes: Based on the degree of influence of different predictive factors on the risk of stillbirth related to umbilical cord torsion, different weight levels are set for the core predictive factor and the auxiliary predictive factor; wherein, the weight level of the core predictive factor is higher than that of the auxiliary predictive factor, and the weight level of the information on changes in fetal movement in pregnant women is higher than that of signs of umbilical cord torsion, and the weight level of signs of umbilical cord torsion is higher than that of information on fetal growth restriction. During the feature vector construction process, the weight level is encoded as a feature parameter into the feature vector.

6. The method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion according to claim 1, characterized in that, The step of inputting the feature vector into a pre-trained risk prediction model and calculating a comprehensive risk score for umbilical cord torsion-related stillbirths based on the risk prediction model includes: The model structure for constructing the risk prediction model includes at least: a feature input layer for receiving the feature vector composed of the core predictive factor and the auxiliary predictive factor; a feature weighting layer for weighting each feature component in the feature vector according to predetermined model coefficients; a risk calculation layer for performing linear combination operations on the weighted feature results and generating intermediate risk values; and a probability mapping layer for performing nonlinear mapping on the intermediate risk values ​​and outputting the predicted probability value of umbilical cord torsion-related stillbirths. The feature vector is input into the feature input layer, and in the feature weighting layer, the core predictive factor is assigned a higher weight coefficient than the auxiliary predictive factor according to the weight level of different predictive factors. In the risk calculation layer, the weighted feature components are summed to obtain the intermediate risk value that characterizes the intensity of the stillbirth risk. In the probability mapping layer, the intermediate risk value is transformed based on a preset probability function to obtain the predicted probability value of the stillbirth related to umbilical cord torsion. Based on the preset probability and scoring mapping rules, the predicted probability value is converted into the comprehensive risk score.

7. The method for predicting and intervening in the risk of stillbirth related to umbilical cord torsion according to claim 1, characterized in that, The step involves automatically matching a corresponding intervention decision-making plan from a pre-set clinical intervention strategy library based on the risk level, and outputting risk assessment results and intervention recommendations to guide clinical treatment, including: When the comprehensive risk score is higher than the first preset threshold, it is determined to be a high-risk level, and an intervention decision plan including immediate hospital admission, continuous fetal heart rate monitoring and preparation for emergency termination of pregnancy is automatically matched. When the comprehensive risk score is between the first preset threshold and the second preset threshold, it is determined to be of medium risk level, and an intervention decision plan including shortening the re-examination cycle, increasing the frequency of prenatal monitoring and implementing further functional assessment is automatically matched. When the comprehensive risk score is lower than the second preset threshold, it is determined to be a low-risk level, and an intervention decision plan including routine prenatal check-up management and enhanced guidance for pregnant women's self-monitoring is automatically matched; The risk level, the comprehensive risk score, and the intervention decision plan are output to the clinical terminal in a visual or structured form.

8. A decision-making system for predicting and intervening in the risk of stillbirth related to umbilical cord torsion, characterized in that, include: An acquisition unit is used to acquire multimodal clinical data of pregnant women and to standardize the multimodal clinical data, wherein the multimodal clinical data includes at least the pregnant woman's chief complaint information and prenatal ultrasound examination information. The extraction unit is used to extract core and auxiliary predictive factors for predicting the risk of stillbirth related to umbilical cord torsion from standardized multimodal clinical data. An encoding unit is used to encode the core predictor and the auxiliary predictor into a feature vector that the model can process; The input unit is used to input the feature vector into a pre-trained risk prediction model and calculate a comprehensive risk score for stillbirths related to umbilical cord torsion based on the risk prediction model. A grading unit is used to classify the comprehensive risk score into risk levels based on a preset risk threshold. The matching unit is used to automatically match the corresponding intervention decision plan from the preset clinical intervention strategy library according to the risk level, and output the risk assessment results and intervention suggestions to guide clinical treatment.

9. An electronic device, comprising: The memory and processor are characterized in that the processor is used to execute a computer program stored in the memory to implement the steps of the prediction and intervention decision-making method for umbilical cord torsion-related stillbirth risk as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the prediction and intervention decision-making method for the risk of stillbirth related to umbilical cord torsion as described in any one of claims 1-7.

Images