技术领域Technical Field
本申请涉及医疗信息处理技术领域，且更为具体地，涉及一种妊娠风险预警管理系统。The present application relates to the technical field of medical information processing, and more specifically, to a pregnancy risk warning management system.
背景技术Background Art
妊娠期是女性生理历程中独特且至关重要的阶段，妊娠期间，母体的生理机能会发生显著的变化，这些变化可能催生一系列妊娠并发症，对母婴的健康造成潜在风险。近年来，随着全球生育观念的演变，高龄妊娠的比率逐年攀升，这在一定程度上加剧了妊娠并发症的可能。因此，对妊娠并发症的早期识别和高效管理成为了保障母婴健康的关键任务。Pregnancy is a unique and crucial stage in a woman's physiological history. During pregnancy, the mother's physiological functions will undergo significant changes, which may give rise to a series of pregnancy complications and pose potential risks to the health of mother and baby. In recent years, with the evolution of global fertility concepts, the rate of advanced pregnancy has increased year by year, which to some extent exacerbates the possibility of pregnancy complications. Therefore, early identification and efficient management of pregnancy complications have become key tasks to ensure the health of mothers and babies.
传统的妊娠管理策略主要基于定期的产科门诊就诊，然而，这种方法可能无法实现对孕妇生理状态的实时监控，可能导致检查不及时、监测连续性差、漏诊率高等问题，其有效性受到一定的制约。Traditional pregnancy management strategies are mainly based on regular obstetric outpatient visits. However, this method may not be able to achieve real-time monitoring of the physiological state of pregnant women, which may lead to problems such as untimely examinations, poor monitoring continuity, and high missed diagnosis rates, and its effectiveness is subject to certain constraints.
随着科技的发展，可穿戴设备和远程监测技术在妇产科领域的应用越来越广泛，使得实时监测孕妇的生理状态成为可能。因此，期待一种智能化的妊娠风险预警管理系统。With the development of science and technology, wearable devices and remote monitoring technology are increasingly used in the field of obstetrics and gynecology, making it possible to monitor the physiological state of pregnant women in real time. Therefore, an intelligent pregnancy risk warning management system is expected.
发明内容Summary of the invention
为了解决上述技术问题，提出了本申请。In order to solve the above technical problems, this application is proposed.
相应地，根据本申请的一个方面，提供了一种妊娠风险预警管理系统，其包括：Accordingly, according to one aspect of the present application, a pregnancy risk warning management system is provided, comprising:
孕妇生理状态监测模块，用于获取由可穿戴设备采集的被监控孕妇对象的生理状态参数的时间序列，其中，所述生理状态参数包括心率值、血压值、脉搏频率值；A pregnant woman physiological state monitoring module, used to obtain a time series of physiological state parameters of the monitored pregnant woman collected by a wearable device, wherein the physiological state parameters include heart rate value, blood pressure value, and pulse frequency value;
生理状态参数时序编码模块，用于对所述生理状态参数的时间序列进行参数样本级时序特征提取以得到心率值时序关联隐含特征向量、血压值时序关联隐含特征向量和脉搏频率值时序关联隐含特征向量；A physiological state parameter time series encoding module, used for performing parameter sample level time series feature extraction on the time series of the physiological state parameter to obtain a heart rate value time series associated implicit feature vector, a blood pressure value time series associated implicit feature vector and a pulse frequency value time series associated implicit feature vector;
生理状态参数规整模块，用于对所述生理状态参数的时间序列进行数据规整以得到心率值时序输入向量、血压值时序输入向量和脉搏频率值时序输入向量；A physiological state parameter regularization module, used for performing data regularization on the time series of the physiological state parameters to obtain a heart rate value time series input vector, a blood pressure value time series input vector and a pulse frequency value time series input vector;
多尺度特征门控融合模块，用于将所述心率值时序关联隐含特征向量和所述心率值时序输入向量、所述血压值时序关联隐含特征向量和所述血压值时序输入向量以及所述脉搏频率值时序关联隐含特征向量和所述脉搏频率值时序输入向量输入基于门控响应的特征动态交互融合模块以得到心率多尺度门控融合特征向量、血压多尺度门控融合特征向量和脉搏频率多尺度门控融合特征向量；A multi-scale feature gated fusion module, used for inputting the heart rate value time series associated implicit feature vector and the heart rate value time series input vector, the blood pressure value time series associated implicit feature vector and the blood pressure value time series input vector, and the pulse frequency value time series associated implicit feature vector and the pulse frequency value time series input vector into a feature dynamic interactive fusion module based on gated response to obtain a heart rate multi-scale gated fusion feature vector, a blood pressure multi-scale gated fusion feature vector, and a pulse frequency multi-scale gated fusion feature vector;
生理状态预警管理模块，用于基于所述血压多尺度门控融合特征向量、所述心率多尺度门控融合特征向量和所述脉搏频率多尺度门控融合特征向量的多模态融合特征，判断所述被监控孕妇对象的生理状态是否存在异常；a physiological state early warning management module, for judging whether the physiological state of the monitored pregnant woman is abnormal based on the multimodal fusion features of the blood pressure multi-scale gated fusion feature vector, the heart rate multi-scale gated fusion feature vector and the pulse frequency multi-scale gated fusion feature vector;
其中，所述多尺度特征门控融合模块，包括：特征联合单元，用于将所述心率值时序关联隐含特征向量和所述心率值时序输入向量输入特征联合模块以得到心率多尺度时序特征联合表示向量；信息融合门控响应计算单元，用于将所述心率多尺度时序特征联合表示向量输入门控响应函数以得到信息融合的响应门；信息融合单元，用于计算一与所述信息融合的响应门的差值，并以所述信息融合的响应门和所述差值作为权重，来计算所述心率值时序关联隐含特征向量和所述心率值时序输入向量的按位置加权和以得到所述心率多尺度门控融合特征向量。Among them, the multi-scale feature gated fusion module includes: a feature combination unit, which is used to input the heart rate value time series associated implicit feature vector and the heart rate value time series input vector into the feature combination module to obtain a heart rate multi-scale time series feature joint representation vector; an information fusion gated response calculation unit, which is used to input the heart rate multi-scale time series feature joint representation vector into the gated response function to obtain an information fusion response gate; an information fusion unit, which is used to calculate a difference with the information fusion response gate, and use the information fusion response gate and the difference as weights to calculate the position-weighted sum of the heart rate value time series associated implicit feature vector and the heart rate value time series input vector to obtain the heart rate multi-scale gated fusion feature vector.
在上述妊娠风险预警管理系统中，所述生理状态参数时序编码模块，包括：参数样本级数据排列单元，用于将所述生理状态参数的时间序列按照参数样本维度进行排列以得到心率值的时间序列、血压值的时间序列和脉搏频率值的时间序列；时序特征提取单元，用于将所述心率值的时间序列、所述血压值的时间序列和所述脉搏频率值的时间序列输入基于双向门控循环单元的序列编码器以得到所述心率值时序关联隐含特征向量、所述血压值时序关联隐含特征向量和所述脉搏频率值时序关联隐含特征向量。In the above-mentioned pregnancy risk warning management system, the physiological state parameter time series encoding module includes: a parameter sample level data arrangement unit, which is used to arrange the time series of the physiological state parameters according to the parameter sample dimension to obtain the time series of heart rate values, the time series of blood pressure values and the time series of pulse frequency values; a time series feature extraction unit, which is used to input the time series of heart rate values, the time series of blood pressure values and the time series of pulse frequency values into a sequence encoder based on a bidirectional gated cyclic unit to obtain the heart rate value time series associated implicit feature vector, the blood pressure value time series associated implicit feature vector and the pulse frequency value time series associated implicit feature vector.
在上述妊娠风险预警管理系统中，所述生理状态参数规整模块，用于：将所述生理状态参数的时间序列按照参数样本维度和时间维度进行一维排列以得到所述心率值时序输入向量、所述血压值时序输入向量和所述脉搏频率值时序输入向量。In the above-mentioned pregnancy risk warning management system, the physiological state parameter regularization module is used to: arrange the time series of the physiological state parameters in one dimension according to the parameter sample dimension and the time dimension to obtain the heart rate value timing input vector, the blood pressure value timing input vector and the pulse frequency value timing input vector.
在上述妊娠风险预警管理系统中，所述特征联合单元，用于：将所述心率值时序关联隐含特征向量和所述心率值时序输入向量进行特征级联以得到所述心率多尺度时序特征联合表示向量。In the above-mentioned pregnancy risk warning management system, the feature combination unit is used to: perform feature cascading on the heart rate value time series associated implicit feature vector and the heart rate value time series input vector to obtain the heart rate multi-scale time series feature joint representation vector.
在上述妊娠风险预警管理系统中，所述信息融合门控响应计算单元，用于：使用预定权重向量乘以所述心率多尺度时序特征联合表示向量以得到信息交互融合相关系数；将所述信息交互融合相关系数和预定偏置参数相加后通过sigmoid函数进行激活处理以得到所述信息融合的响应门。In the above-mentioned pregnancy risk warning management system, the information fusion gated response calculation unit is used to: use a predetermined weight vector to multiply the heart rate multi-scale time series feature joint representation vector to obtain the information interaction fusion correlation coefficient; add the information interaction fusion correlation coefficient and the predetermined bias parameter and then activate it through a sigmoid function to obtain the response gate of the information fusion.
在上述妊娠风险预警管理系统中，所述生理状态预警管理模块，包括：多模态特征融合单元，用于将所述血压多尺度门控融合特征向量、所述心率多尺度门控融合特征向量和所述脉搏频率多尺度门控融合特征向量输入基于嵌入层的多模态特征融合器以得到生理状态多模态融合特征向量；预警管理结果生成单元，用于将所述生理状态多模态融合特征向量输入基于分类器的预警管理结果生成器以得到预警管理结果，所述预警管理结果用于表示所述被监控孕妇对象的生理状态是否存在异常。In the above-mentioned pregnancy risk warning management system, the physiological state warning management module includes: a multimodal feature fusion unit, which is used to input the blood pressure multi-scale gated fusion feature vector, the heart rate multi-scale gated fusion feature vector and the pulse frequency multi-scale gated fusion feature vector into a multimodal feature fuser based on an embedding layer to obtain a physiological state multimodal fusion feature vector; a warning management result generation unit, which is used to input the physiological state multimodal fusion feature vector into a classifier-based warning management result generator to obtain a warning management result, and the warning management result is used to indicate whether the physiological state of the monitored pregnant woman is abnormal.
在上述妊娠风险预警管理系统中，还包括用于对所述基于双向门控循环单元的序列编码器、所述基于门控响应的特征动态交互融合模块、所述基于嵌入层的多模态特征融合器和所述基于分类器的预警管理结果生成器进行训练的训练模块。In the above-mentioned pregnancy risk warning management system, it also includes a training module for training the sequence encoder based on the bidirectional gated recurrent unit, the feature dynamic interaction fusion module based on the gated response, the multimodal feature fuser based on the embedding layer and the classifier-based warning management result generator.
在上述妊娠风险预警管理系统中，所述训练模块，包括：训练数据获取单元，用于获取训练数据，所述训练数据包括由可穿戴设备采集的被监控孕妇对象的训练生理状态参数的时间序列，其中，所述训练生理状态参数包括训练心率值、训练血压值、训练脉搏频率值以及所述被监控孕妇对象的生理状态是否存在异常的真实值；训练参数样本维度排列单元，用于将所述训练生理状态参数的时间序列按照参数样本维度进行排列以得到训练心率值的时间序列、训练血压值的时间序列和训练脉搏频率值的时间序列；训练参数时序编码单元，用于将所述训练心率值的时间序列、所述训练血压值的时间序列和所述训练脉搏频率值的时间序列输入所述基于双向门控循环单元的序列编码器以得到训练心率值时序关联隐含特征向量、训练血压值时序关联隐含特征向量和训练脉搏频率值时序关联隐含特征向量；训练参数时序排列单元，用于将所述训练心率值的时间序列、所述训练血压值的时间序列和所述训练脉搏频率值的时间序列分别按照时间维度排列为训练心率值时序输入向量、训练血压值时序输入向量和训练脉搏频率值时序输入向量；训练参数多尺度时序特征融合单元，用于将所述训练心率值时序关联隐含特征向量和所述训练心率值时序输入向量、所述训练血压值时序关联隐含特征向量和所述训练血压值时序输入向量以及所述训练脉搏频率值时序关联隐含特征向量和所述训练脉搏频率值时序输入向量输入所述基于门控响应的特征动态交互融合模块以得到训练心率多尺度门控融合特征向量、训练血压多尺度门控融合特征向量和训练脉搏频率多尺度门控融合特征向量；训练参数多模态特征融合单元，用于将所述训练血压多尺度门控融合特征向量、所述训练心率多尺度门控融合特征向量和所述训练脉搏频率多尺度门控融合特征向量输入所述基于嵌入层的多模态特征融合器以得到训练生理状态多模态融合特征向量；分类损失单元，用于将所述训练生理状态多模态融合特征向量输入所述基于分类器的预警管理结果生成器以得到分类损失函数值；预定损失函数值计算单元，用于计算所述训练生理状态多模态融合特征向量的预定损失函数值；模型训练单元，用于以所述分类损失函数值和所述预定损失函数值的加权和作为损失函数值，来对所述基于双向门控循环单元的序列编码器、所述基于门控响应的特征动态交互融合模块、所述基于嵌入层的多模态特征融合器和所述基于分类器的预警管理结果生成器进行训练。In the above-mentioned pregnancy risk warning management system, the training module includes: a training data acquisition unit, which is used to acquire training data, wherein the training data includes a time series of training physiological state parameters of the monitored pregnant woman collected by the wearable device, wherein the training physiological state parameters include training heart rate values, training blood pressure values, training pulse frequency values, and a real value of whether the physiological state of the monitored pregnant woman is abnormal; a training parameter sample dimension arrangement unit, which is used to arrange the time series of the training physiological state parameters according to the parameter sample dimension to obtain a time series of training heart rate values, a time series of training blood pressure values, and a time series of training pulse frequency values; a training parameter timing encoding unit, which is used to encode the training heart rate values. The time series of the training heart rate values, the time series of the training blood pressure values, and the time series of the training pulse frequency values are input into the sequence encoder based on the bidirectional gated cyclic unit to obtain the training heart rate value time series associated implicit feature vector, the training blood pressure value time series associated implicit feature vector, and the training pulse frequency value time series associated implicit feature vector; a training parameter time series arrangement unit is used to arrange the time series of the training heart rate values, the time series of the training blood pressure values, and the time series of the training pulse frequency values according to the time dimension as the training heart rate value time series input vector, the training blood pressure value time series input vector, and the training pulse frequency value time series input vector respectively; a training parameter multi-scale time series feature fusion unit is used to combine the training heart rate value time series associated implicit feature vector The vector and the training heart rate value time series input vector, the training blood pressure value time series associated implicit feature vector and the training blood pressure value time series input vector, and the training pulse frequency value time series associated implicit feature vector and the training pulse frequency value time series input vector are input into the feature dynamic interactive fusion module based on gated response to obtain the training heart rate multi-scale gated fusion feature vector, the training blood pressure multi-scale gated fusion feature vector and the training pulse frequency multi-scale gated fusion feature vector; the training parameter multimodal feature fusion unit is used to input the training blood pressure multi-scale gated fusion feature vector, the training heart rate multi-scale gated fusion feature vector and the training pulse frequency multi-scale gated fusion feature vector into the multi-modal based on embedding layer. A feature fusion device is used to obtain a training physiological state multimodal fusion feature vector; a classification loss unit is used to input the training physiological state multimodal fusion feature vector into the classifier-based early warning management result generator to obtain a classification loss function value; a predetermined loss function value calculation unit is used to calculate the predetermined loss function value of the training physiological state multimodal fusion feature vector; a model training unit is used to train the sequence encoder based on the bidirectional gated recurrent unit, the feature dynamic interaction fusion module based on the gated response, the multimodal feature fusion device based on the embedding layer, and the classifier-based early warning management result generator using the weighted sum of the classification loss function value and the predetermined loss function value as the loss function value.
与现有技术相比，本申请提供的妊娠风险预警管理系统采用基于深度学习的人工智能技术对孕妇的生理状态进行实时监测和数据分析，以捕捉到孕妇的心率、血压和脉搏频率的时序变化特性，进而基于孕妇各项生理状态参数的时序多模态融合特征来智能判断孕妇的生理状态是否存在异常。这样，减轻医疗人员的工作负担，保障母婴的健康安全。Compared with the existing technology, the pregnancy risk warning management system provided by this application uses artificial intelligence technology based on deep learning to monitor and analyze the physiological state of pregnant women in real time, so as to capture the time series change characteristics of the heart rate, blood pressure and pulse frequency of pregnant women, and then intelligently judge whether the physiological state of pregnant women is abnormal based on the time series multimodal fusion characteristics of various physiological state parameters of pregnant women. In this way, the workload of medical personnel is reduced and the health and safety of mothers and babies are guaranteed.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
通过结合附图对本申请实施例进行更详细的描述，本申请的上述以及其他目的、特征和优势将变得更加明显。附图用来提供对本申请实施例的进一步理解，并且构成说明书的一部分，与本申请实施例一起用于解释本申请，并不构成对本申请的限制。在附图中，相同的参考标号通常代表相同部件或步骤。By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.
图1为根据本申请实施例的妊娠风险预警管理系统的框图。FIG1 is a block diagram of a pregnancy risk warning management system according to an embodiment of the present application.
图2为根据本申请实施例的妊娠风险预警管理系统的架构示意图。FIG2 is a schematic diagram of the architecture of a pregnancy risk warning management system according to an embodiment of the present application.
图3为根据本申请实施例的妊娠风险预警管理系统中生理状态参数时序编码模块的框图。FIG3 is a block diagram of a physiological state parameter time series encoding module in a pregnancy risk warning management system according to an embodiment of the present application.
图4为根据本申请实施例的妊娠风险预警管理系统中多尺度特征门控融合模块的框图。FIG4 is a block diagram of a multi-scale feature gating fusion module in a pregnancy risk warning management system according to an embodiment of the present application.
图5为根据本申请实施例的妊娠风险预警管理系统中生理状态预警管理模块的框图。FIG5 is a block diagram of a physiological state warning management module in a pregnancy risk warning management system according to an embodiment of the present application.
图6为根据本申请实施例的妊娠风险预警管理系统中训练模块的框图。FIG6 is a block diagram of a training module in a pregnancy risk warning management system according to an embodiment of the present application.
具体实施方式DETAILED DESCRIPTION
下面，将结合附图对本申请实施例进行更详细的描述，本申请的上述以及其他目的、特征和优势将变得更加明显。显然，所描述的实施例仅仅是本申请的一部分实施例，而不是本申请的全部实施例，应理解，本申请不受这里描述的示例实施例的限制。Below, the embodiments of the present application will be described in more detail in conjunction with the accompanying drawings, and the above and other purposes, features and advantages of the present application will become more apparent. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein.
如上述背景技术所言，传统的妊娠管理策略主要基于定期的产科门诊就诊，然而，这种方法可能无法实现对孕妇生理状态的实时监控，可能导致检查不及时、监测连续性差、漏诊率高等问题，其有效性受到一定的制约。针对上述技术问题，本申请的技术构思为采用基于深度学习的人工智能技术对孕妇的生理状态进行实时监测和数据分析，以捕捉到孕妇的心率、血压和脉搏频率的时序变化特性，进而基于孕妇各项生理状态参数的时序多模态融合特征来智能判断孕妇的生理状态是否存在异常。这样，减轻医疗人员的工作负担，保障母婴的健康安全。As mentioned in the above background technology, traditional pregnancy management strategies are mainly based on regular obstetric outpatient visits. However, this method may not be able to achieve real-time monitoring of the physiological state of pregnant women, which may lead to problems such as untimely examinations, poor monitoring continuity, and high missed diagnosis rates. Its effectiveness is subject to certain restrictions. In response to the above technical problems, the technical concept of this application is to use artificial intelligence technology based on deep learning to conduct real-time monitoring and data analysis of the physiological state of pregnant women, so as to capture the time series variation characteristics of the heart rate, blood pressure and pulse frequency of pregnant women, and then intelligently judge whether there is any abnormality in the physiological state of pregnant women based on the time series multimodal fusion characteristics of various physiological state parameters of pregnant women. In this way, the workload of medical personnel is reduced and the health and safety of mothers and babies are guaranteed.
图1为根据本申请实施例的妊娠风险预警管理系统的框图。图2为根据本申请实施例的妊娠风险预警管理系统的架构示意图。如图1和图2所示，根据本申请实施例的妊娠风险预警管理系统100，包括：孕妇生理状态监测模块110，用于获取由可穿戴设备采集的被监控孕妇对象的生理状态参数的时间序列，其中，所述生理状态参数包括心率值、血压值、脉搏频率值；生理状态参数时序编码模块120，用于对所述生理状态参数的时间序列进行参数样本级时序特征提取以得到心率值时序关联隐含特征向量、血压值时序关联隐含特征向量和脉搏频率值时序关联隐含特征向量；生理状态参数规整模块130，用于对所述生理状态参数的时间序列进行数据规整以得到心率值时序输入向量、血压值时序输入向量和脉搏频率值时序输入向量；多尺度特征门控融合模块140，用于将所述心率值时序关联隐含特征向量和所述心率值时序输入向量、所述血压值时序关联隐含特征向量和所述血压值时序输入向量以及所述脉搏频率值时序关联隐含特征向量和所述脉搏频率值时序输入向量输入基于门控响应的特征动态交互融合模块以得到心率多尺度门控融合特征向量、血压多尺度门控融合特征向量和脉搏频率多尺度门控融合特征向量；生理状态预警管理模块150，用于基于所述血压多尺度门控融合特征向量、所述心率多尺度门控融合特征向量和所述脉搏频率多尺度门控融合特征向量的多模态融合特征，判断所述被监控孕妇对象的生理状态是否存在异常。FIG1 is a block diagram of a pregnancy risk warning management system according to an embodiment of the present application. FIG2 is a schematic diagram of the architecture of a pregnancy risk warning management system according to an embodiment of the present application. As shown in FIG1 and FIG2, a pregnancy risk warning management system 100 according to an embodiment of the present application includes: a pregnant woman physiological state monitoring module 110, which is used to obtain a time series of physiological state parameters of a monitored pregnant woman collected by a wearable device, wherein the physiological state parameters include heart rate values, blood pressure values, and pulse frequency values; a physiological state parameter timing encoding module 120, which is used to perform parameter sample-level timing feature extraction on the time series of the physiological state parameters to obtain a heart rate value timing-related implicit feature vector, a blood pressure value timing-related implicit feature vector, and a pulse frequency value timing-related implicit feature vector; a physiological state parameter regularization module 130, which is used to perform data regularization on the time series of the physiological state parameters to obtain a heart rate value timing input vector, a blood pressure value timing input vector, and a pulse frequency value timing input vector ; A multi-scale feature gated fusion module 140, used to input the heart rate value time series associated implicit feature vector and the heart rate value time series input vector, the blood pressure value time series associated implicit feature vector and the blood pressure value time series input vector, and the pulse frequency value time series associated implicit feature vector and the pulse frequency value time series input vector into a feature dynamic interactive fusion module based on gated response to obtain a heart rate multi-scale gated fusion feature vector, a blood pressure multi-scale gated fusion feature vector, and a pulse frequency multi-scale gated fusion feature vector; A physiological state early warning management module 150, used to determine whether the physiological state of the monitored pregnant woman is abnormal based on the multimodal fusion features of the blood pressure multi-scale gated fusion feature vector, the heart rate multi-scale gated fusion feature vector, and the pulse frequency multi-scale gated fusion feature vector.
在上述妊娠风险预警管理系统100中，所述孕妇生理状态监测模块110，用于获取由可穿戴设备采集的被监控孕妇对象的生理状态参数的时间序列，其中，所述生理状态参数包括心率值、血压值、脉搏频率值。应可以理解，孕妇在妊娠期间，其身体机能会经历显著的变化，包括心脏负担加重、血容量增加等。这些身体机能的变化可能导致其心率、血压、脉搏频率等生理参数的变化。因此，通过实时监测孕妇的生理状态参数，并对孕妇的生理状态参数进行深入的时序分析，可以挖掘出孕妇的生理状态随时间变化的趋势和规律，有助于更准确地判断孕妇的生理状态是否存在异常。相比较现有的一些基于预警阈值的妊娠风险预警系统，在本申请的技术方案中，更注重生理参数的动态变化模式，有效提高了预警的准确性。例如，短暂的心率加快可能是正常的生理反应，但持续的心率加快或异常的血压波动可能提示孕妇存在潜在的健康问题，如妊娠期高血压疾病或心脏病等。In the above-mentioned pregnancy risk warning management system 100, the pregnant woman physiological state monitoring module 110 is used to obtain the time series of physiological state parameters of the monitored pregnant woman collected by the wearable device, wherein the physiological state parameters include heart rate value, blood pressure value, and pulse frequency value. It should be understood that during pregnancy, the body functions of pregnant women will undergo significant changes, including increased heart burden, increased blood volume, etc. These changes in body functions may cause changes in physiological parameters such as heart rate, blood pressure, and pulse frequency. Therefore, by real-time monitoring of the physiological state parameters of pregnant women and conducting in-depth time series analysis of the physiological state parameters of pregnant women, the trend and law of changes in the physiological state of pregnant women over time can be excavated, which helps to more accurately determine whether the physiological state of pregnant women is abnormal. Compared with some existing pregnancy risk warning systems based on warning thresholds, in the technical solution of this application, more attention is paid to the dynamic change mode of physiological parameters, which effectively improves the accuracy of warning. For example, a short-term increase in heart rate may be a normal physiological reaction, but a continuous increase in heart rate or abnormal blood pressure fluctuations may indicate that pregnant women have potential health problems, such as gestational hypertension or heart disease.
在上述妊娠风险预警管理系统100中，所述生理状态参数时序编码模块120，用于对所述生理状态参数的时间序列进行参数样本级时序特征提取以得到心率值时序关联隐含特征向量、血压值时序关联隐含特征向量和脉搏频率值时序关联隐含特征向量。其中，图3为根据本申请实施例的妊娠风险预警管理系统中生理状态参数时序编码模块的框图。如图3所示，所述生理状态参数时序编码模块120，包括：参数样本级数据排列单元121，用于将所述生理状态参数的时间序列按照参数样本维度进行排列以得到心率值的时间序列、血压值的时间序列和脉搏频率值的时间序列；时序特征提取单元122，用于将所述心率值的时间序列、所述血压值的时间序列和所述脉搏频率值的时间序列输入基于双向门控循环单元的序列编码器以得到所述心率值时序关联隐含特征向量、所述血压值时序关联隐含特征向量和所述脉搏频率值时序关联隐含特征向量。In the above-mentioned pregnancy risk warning management system 100, the physiological state parameter time series encoding module 120 is used to extract parameter sample level time series features of the time series of the physiological state parameters to obtain the heart rate value time series associated implicit feature vector, the blood pressure value time series associated implicit feature vector and the pulse frequency value time series associated implicit feature vector. Among them, Figure 3 is a block diagram of the physiological state parameter time series encoding module in the pregnancy risk warning management system according to an embodiment of the present application. As shown in Figure 3, the physiological state parameter time series encoding module 120 includes: a parameter sample level data arrangement unit 121, which is used to arrange the time series of the physiological state parameters according to the parameter sample dimension to obtain the time series of the heart rate value, the time series of the blood pressure value and the time series of the pulse frequency value; a time series feature extraction unit 122, which is used to input the time series of the heart rate value, the time series of the blood pressure value and the time series of the pulse frequency value into a sequence encoder based on a bidirectional gated cyclic unit to obtain the heart rate value time series associated implicit feature vector, the blood pressure value time series associated implicit feature vector and the pulse frequency value time series associated implicit feature vector.
具体地，所述参数样本级数据排列单元121，用于将所述生理状态参数的时间序列按照参数样本维度进行排列以得到心率值的时间序列、血压值的时间序列和脉搏频率值的时间序列。应可以理解，考虑到不同的生理状态参数反映了孕妇身体不同方面的健康状况，因此，为了更精确地捕捉各项参数的时序变化趋势，在本申请的技术方案中，进一步将所述生理状态参数的时间序列按照参数样本维度进行排列，以将不同类型的生理状态参数（如心率、血压、脉搏频率）分别整理成独立的时间序列，使得数据更加清晰、易于管理和分析，有助于在后续处理过程中减少混淆，更敏感地捕捉到各项参数的异常变化，提高数据处理的效率和准确性。Specifically, the parameter sample level data arrangement unit 121 is used to arrange the time series of the physiological state parameters according to the parameter sample dimension to obtain the time series of the heart rate value, the time series of the blood pressure value and the time series of the pulse frequency value. It should be understood that considering that different physiological state parameters reflect the health status of pregnant women in different aspects, therefore, in order to more accurately capture the time series change trend of each parameter, in the technical solution of the present application, the time series of the physiological state parameters are further arranged according to the parameter sample dimension, so that different types of physiological state parameters (such as heart rate, blood pressure, pulse frequency) are sorted into independent time series, so that the data is clearer, easier to manage and analyze, which helps to reduce confusion in the subsequent processing process, more sensitively capture abnormal changes in various parameters, and improve the efficiency and accuracy of data processing.
具体地，所述时序特征提取单元122，用于将所述心率值的时间序列、所述血压值的时间序列和所述脉搏频率值的时间序列输入基于双向门控循环单元的序列编码器以得到所述心率值时序关联隐含特征向量、所述血压值时序关联隐含特征向量和所述脉搏频率值时序关联隐含特征向量。也就是，为了有效捕捉到孕妇的各项生理状态参数在时间维度上复杂的时序依赖关系，在本申请的技术方案中，采用了基于双向门控循环单元（Bi-GRU）的序列编码器对各项生理状态参数的时间序列数据分别进行时序编码。本领域普通技术人员应知晓，传统的RNN在处理长序列时容易出现梯度消失或梯度爆炸的问题，导致无法有效学习数据中的长期依赖关系。而Bi-GRU模型作为一种特殊的循环神经网络（RNN），通过引入更新门和重置门，有效地缓解了这一问题，使得模型能够保留长期记忆，捕捉到各项生理状态参数的时间序列数据中的长期依赖性。同时，Bi-GRU模型的双向结构则可以同时捕获时间序列数据中前向和后向的依赖关系，从而能够更全面地理解孕妇生理状态参数的变化趋势，如周期性变化、异常突变等，为后续的生理状态异常检测提供强有力的支持。Specifically, the time series feature extraction unit 122 is used to input the time series of the heart rate value, the time series of the blood pressure value and the time series of the pulse frequency value into a sequence encoder based on a bidirectional gated recurrent unit to obtain the heart rate value time series associated implicit feature vector, the blood pressure value time series associated implicit feature vector and the pulse frequency value time series associated implicit feature vector. That is, in order to effectively capture the complex time series dependencies of various physiological state parameters of pregnant women in the time dimension, in the technical solution of the present application, a sequence encoder based on a bidirectional gated recurrent unit (Bi-GRU) is used to perform time series encoding on the time series data of various physiological state parameters. It should be known to those skilled in the art that traditional RNNs are prone to gradient vanishing or gradient explosion problems when processing long sequences, resulting in the inability to effectively learn long-term dependencies in the data. As a special recurrent neural network (RNN), the Bi-GRU model effectively alleviates this problem by introducing update gates and reset gates, so that the model can retain long-term memory and capture long-term dependencies in the time series data of various physiological state parameters. At the same time, the bidirectional structure of the Bi-GRU model can simultaneously capture the forward and backward dependencies in the time series data, so as to more comprehensively understand the changing trends of the physiological state parameters of pregnant women, such as periodic changes, abnormal mutations, etc., and provide strong support for the subsequent detection of physiological abnormalities.
在上述妊娠风险预警管理系统100中，所述生理状态参数规整模块130，用于对所述生理状态参数的时间序列进行数据规整以得到心率值时序输入向量、血压值时序输入向量和脉搏频率值时序输入向量。在本申请的一个具体示例中，对所述生理状态参数的时间序列进行数据规整的处理方式是将所述生理状态参数的时间序列按照参数样本维度和时间维度进行一维排列以得到所述心率值时序输入向量、所述血压值时序输入向量和所述脉搏频率值时序输入向量。应可以理解，考虑到在心率值、血压值、脉搏频率值的时序编码过程中，双向门控循环单元可能更侧重于各项参数的长期时序关联趋势，而忽略了一些局部的、短时尺度的参数变化。因此，为了增强模型对各项生理状态参数时序数据的多尺度理解，本申请的技术方案中，进一步将孕妇的各项生理状态参数的时间序列数据按照时间顺序排列为一维向量形式，以保留各项生理状态参数原始数据的时间顺序关系，获取各项参数的源时序分布特征，为孕妇生理状态参数的动态分析提供更丰富的时序信息。In the above-mentioned pregnancy risk warning management system 100, the physiological state parameter regularization module 130 is used to perform data regularization on the time series of the physiological state parameters to obtain the heart rate value timing input vector, the blood pressure value timing input vector and the pulse frequency value timing input vector. In a specific example of the present application, the processing method for performing data regularization on the time series of the physiological state parameters is to arrange the time series of the physiological state parameters in one dimension according to the parameter sample dimension and the time dimension to obtain the heart rate value timing input vector, the blood pressure value timing input vector and the pulse frequency value timing input vector. It should be understood that, considering that in the timing encoding process of the heart rate value, the blood pressure value and the pulse frequency value, the bidirectional gated recurrent unit may focus more on the long-term timing correlation trend of each parameter, while ignoring some local, short-time scale parameter changes. Therefore, in order to enhance the model's multi-scale understanding of the time series data of various physiological state parameters, in the technical solution of the present application, the time series data of various physiological state parameters of pregnant women are further arranged in chronological order into a one-dimensional vector form to retain the time sequence relationship of the original data of various physiological state parameters, obtain the source time series distribution characteristics of various parameters, and provide richer time series information for the dynamic analysis of the physiological state parameters of pregnant women.
在上述妊娠风险预警管理系统100中，所述多尺度特征门控融合模块140，用于将所述心率值时序关联隐含特征向量和所述心率值时序输入向量、所述血压值时序关联隐含特征向量和所述血压值时序输入向量以及所述脉搏频率值时序关联隐含特征向量和所述脉搏频率值时序输入向量输入基于门控响应的特征动态交互融合模块以得到心率多尺度门控融合特征向量、血压多尺度门控融合特征向量和脉搏频率多尺度门控融合特征向量。应可以理解，为了更全面地捕捉到孕妇各项生理状态参数在不同时间尺度上的变化特征，进一步将各项生理状态参数经过时序编码器得到的时序关联隐含特征向量与原始的时序输入向量进行融合处理。在本申请的技术方案中，采用了基于门控响应的特征动态交互融合模块来实现不同尺度特征的动态交互和融合。具体地，所述特征动态交互融合模块基于门控响应机制，能够通过学习不同尺度特征之间的关联交互关系，动态调整不同尺度信息的融合权重，使得模型能够根据需要更加灵活地关注局部的短期变化或全局的长期趋势，从而实现多尺度特征的深度融合，有助于更准确地识别出可能的异常模式，提高孕妇生理状态异常检测的敏感性和精确度。In the above-mentioned pregnancy risk warning management system 100, the multi-scale feature gating fusion module 140 is used to input the heart rate value time series associated implicit feature vector and the heart rate value time series input vector, the blood pressure value time series associated implicit feature vector and the blood pressure value time series input vector, and the pulse frequency value time series associated implicit feature vector and the pulse frequency value time series input vector into the feature dynamic interactive fusion module based on the gated response to obtain the heart rate multi-scale gated fusion feature vector, the blood pressure multi-scale gated fusion feature vector and the pulse frequency multi-scale gated fusion feature vector. It should be understood that in order to more comprehensively capture the changing characteristics of various physiological state parameters of pregnant women on different time scales, the time series associated implicit feature vectors obtained by the time series encoder of each physiological state parameter are further fused with the original time series input vector. In the technical solution of the present application, a feature dynamic interactive fusion module based on gated response is adopted to realize the dynamic interaction and fusion of features of different scales. Specifically, the feature dynamic interactive fusion module is based on a gated response mechanism and can dynamically adjust the fusion weights of information at different scales by learning the associated interactive relationships between features at different scales, so that the model can more flexibly focus on local short-term changes or global long-term trends as needed, thereby achieving deep fusion of multi-scale features, which helps to more accurately identify possible abnormal patterns and improve the sensitivity and accuracy of detecting abnormalities in the physiological state of pregnant women.
图4为根据本申请实施例的妊娠风险预警管理系统中多尺度特征门控融合模块的框图。如图4所示，所述多尺度特征门控融合模块140，包括：特征联合单元141，用于将所述心率值时序关联隐含特征向量和所述心率值时序输入向量输入特征联合模块以得到心率多尺度时序特征联合表示向量；信息融合门控响应计算单元142，用于将所述心率多尺度时序特征联合表示向量输入门控响应函数以得到信息融合的响应门；信息融合单元143，用于计算一与所述信息融合的响应门的差值，并以所述信息融合的响应门和所述差值作为权重，来计算所述心率值时序关联隐含特征向量和所述心率值时序输入向量的按位置加权和以得到所述心率多尺度门控融合特征向量。Fig. 4 is a block diagram of a multi-scale feature gating fusion module in a pregnancy risk warning management system according to an embodiment of the present application. As shown in Fig. 4, the multi-scale feature gating fusion module 140 includes: a feature combination unit 141, which is used to input the heart rate value time series associated implicit feature vector and the heart rate value time series input vector into the feature combination module to obtain a heart rate multi-scale time series feature joint representation vector; an information fusion gated response calculation unit 142, which is used to input the heart rate multi-scale time series feature joint representation vector into the gated response function to obtain an information fusion response gate; an information fusion unit 143, which is used to calculate a difference with the information fusion response gate, and use the information fusion response gate and the difference as weights to calculate the position-weighted sum of the heart rate value time series associated implicit feature vector and the heart rate value time series input vector to obtain the heart rate multi-scale gated fusion feature vector.
具体地，所述特征联合单元141，用于：将所述心率值时序关联隐含特征向量和所述心率值时序输入向量进行特征级联以得到所述心率多尺度时序特征联合表示向量。Specifically, the feature combining unit 141 is used to perform feature concatenation on the heart rate value time series associated implicit feature vector and the heart rate value time series input vector to obtain the heart rate multi-scale time series feature joint representation vector.
具体地，所述信息融合门控响应计算单元142，用于：使用预定权重向量乘以所述心率多尺度时序特征联合表示向量以得到信息交互融合相关系数；将所述信息交互融合相关系数和预定偏置参数相加后通过sigmoid函数进行激活处理以得到所述信息融合的响应门。Specifically, the information fusion gated response calculation unit 142 is used to: use a predetermined weight vector to multiply the heart rate multi-scale time series feature joint representation vector to obtain an information interaction fusion correlation coefficient; add the information interaction fusion correlation coefficient and a predetermined bias parameter and then activate it through a sigmoid function to obtain a response gate of the information fusion.
也就是，所述多尺度特征门控融合模块140，用于：以如下交互融合公式对所述心率值时序关联隐含特征向量和所述心率值时序输入向量进行融合处理以得到所述心率多尺度门控融合特征向量，其中，所述交互融合公式为；That is, the multi-scale feature gating fusion module 140 is used to: fuse the heart rate value time series associated implicit feature vector and the heart rate value time series input vector using the following interactive fusion formula to obtain the heart rate multi-scale gating fusion feature vector, wherein the interactive fusion formula is:
； ;
； ;
其中，是所述心率值时序关联隐含特征向量，是所述心率值时序输入向量，表示级联操作，是sigmoid函数，是预定权重向量，是预定偏置参数，是信息融合的响应门，是所述心率多尺度门控融合特征向量。in, is the temporal association implicit feature vector of the heart rate value, is the heart rate value time series input vector, Indicates cascade operation, is the sigmoid function, is the predetermined weight vector, is the predetermined bias parameter, is the response gate of information fusion, is the heart rate multi-scale gated fusion feature vector.
在上述妊娠风险预警管理系统100中，所述生理状态预警管理模块150，用于基于所述血压多尺度门控融合特征向量、所述心率多尺度门控融合特征向量和所述脉搏频率多尺度门控融合特征向量的多模态融合特征，判断所述被监控孕妇对象的生理状态是否存在异常。其中，图5为根据本申请实施例的妊娠风险预警管理系统中生理状态预警管理模块的框图。如图5所示，所述生理状态预警管理模块150，包括：多模态特征融合单元151，用于将所述血压多尺度门控融合特征向量、所述心率多尺度门控融合特征向量和所述脉搏频率多尺度门控融合特征向量输入基于嵌入层的多模态特征融合器以得到生理状态多模态融合特征向量；预警管理结果生成单元152，用于将所述生理状态多模态融合特征向量输入基于分类器的预警管理结果生成器以得到预警管理结果，所述预警管理结果用于表示所述被监控孕妇对象的生理状态是否存在异常。In the above-mentioned pregnancy risk warning management system 100, the physiological state warning management module 150 is used to determine whether the physiological state of the monitored pregnant woman is abnormal based on the multimodal fusion features of the blood pressure multi-scale gated fusion feature vector, the heart rate multi-scale gated fusion feature vector and the pulse frequency multi-scale gated fusion feature vector. Among them, Figure 5 is a block diagram of the physiological state warning management module in the pregnancy risk warning management system according to an embodiment of the present application. As shown in Figure 5, the physiological state warning management module 150 includes: a multimodal feature fusion unit 151, which is used to input the blood pressure multi-scale gated fusion feature vector, the heart rate multi-scale gated fusion feature vector and the pulse frequency multi-scale gated fusion feature vector into a multimodal feature fusion device based on an embedding layer to obtain a physiological state multimodal fusion feature vector; a warning management result generation unit 152, which is used to input the physiological state multimodal fusion feature vector into a warning management result generator based on a classifier to obtain a warning management result, and the warning management result is used to indicate whether the physiological state of the monitored pregnant woman is abnormal.
具体地，所述多模态特征融合单元151，用于将所述血压多尺度门控融合特征向量、所述心率多尺度门控融合特征向量和所述脉搏频率多尺度门控融合特征向量输入基于嵌入层的多模态特征融合器以得到生理状态多模态融合特征向量。应可以理解，考虑到心率、血压和脉搏频率分别反映了孕妇的心血管、心肌负荷和呼吸功能等多方面的健康状态，三者在不同生理层面的动态变化可能存在复杂的交叉影响。因此，为了更全面地刻画孕妇整体生理状态的多维度特征，在本申请的技术方案中，进一步引入了基于嵌入层的多模态特征融合器对所述血压多尺度门控融合特征向量、所述心率多尺度门控融合特征向量和所述脉搏频率多尺度门控融合特征向量进行深度整合，以捕捉到不同生理状态参数之间的潜在交互模式和非线性关联。其中，所述多模态特征融合器通过嵌入层的学习能力，能够将不同生理状态参数的多尺度特征向量映射到同一特征空间中，使得来自不同生理层面的信息能够有效地相互作用和互补，增强模型对复杂生理状态变化的建模能力，从而更精确地捕捉到孕妇生理状态的全局特征，提高异常检测的全面性和鲁棒性。Specifically, the multimodal feature fusion unit 151 is used to input the blood pressure multi-scale gated fusion feature vector, the heart rate multi-scale gated fusion feature vector and the pulse frequency multi-scale gated fusion feature vector into a multimodal feature fuser based on an embedding layer to obtain a multimodal fusion feature vector of a physiological state. It should be understood that, considering that heart rate, blood pressure and pulse frequency respectively reflect the health status of pregnant women in many aspects such as cardiovascular, myocardial load and respiratory function, the dynamic changes of the three at different physiological levels may have complex cross-influences. Therefore, in order to more comprehensively characterize the multi-dimensional characteristics of the overall physiological state of pregnant women, in the technical solution of the present application, a multimodal feature fuser based on an embedding layer is further introduced to deeply integrate the blood pressure multi-scale gated fusion feature vector, the heart rate multi-scale gated fusion feature vector and the pulse frequency multi-scale gated fusion feature vector, so as to capture the potential interaction patterns and nonlinear associations between different physiological state parameters. Among them, the multimodal feature fuser can map the multi-scale feature vectors of different physiological state parameters into the same feature space through the learning ability of the embedding layer, so that information from different physiological levels can effectively interact and complement each other, enhance the model's ability to model complex physiological state changes, thereby more accurately capturing the global characteristics of the physiological state of pregnant women and improving the comprehensiveness and robustness of anomaly detection.
具体地，所述预警管理结果生成单元152，用于将所述生理状态多模态融合特征向量输入基于分类器的预警管理结果生成器以得到预警管理结果，所述预警管理结果用于表示所述被监控孕妇对象的生理状态是否存在异常。在本申请的技术方案中，所述基于分类器的预警管理结果生成器用于根据所述生理状态多模态融合特征向量的复杂特征模式，判断被监控的孕妇对象是否存在生理状态异常。在训练阶段，分类器通过大量的标注数据，能够学习到正常生理状态和异常生理状态之间的特征模式差异，形成区分两者的决策边界。在预测阶段，对于输入的所述生理状态多模态融合特征向量，分类器可以基于训练过程学习到的决策规则，将其映射到预定义的类别空间中，以生成对应的预警管理结果，为临床医生提供及时、准确的预警信息，以便于及时采取必要的医疗干预措施，保障孕妇的健康安全。Specifically, the early warning management result generating unit 152 is used to input the multimodal fusion feature vector of the physiological state into the early warning management result generator based on the classifier to obtain the early warning management result, and the early warning management result is used to indicate whether the physiological state of the monitored pregnant woman object is abnormal. In the technical solution of the present application, the early warning management result generator based on the classifier is used to determine whether the monitored pregnant woman object has an abnormal physiological state according to the complex feature pattern of the multimodal fusion feature vector of the physiological state. In the training stage, the classifier can learn the feature pattern difference between the normal physiological state and the abnormal physiological state through a large amount of annotated data, and form a decision boundary to distinguish the two. In the prediction stage, for the input multimodal fusion feature vector of the physiological state, the classifier can map it to the predefined category space based on the decision rules learned in the training process to generate the corresponding early warning management result, and provide timely and accurate early warning information for clinicians, so as to take necessary medical intervention measures in time to ensure the health and safety of pregnant women.
应可以理解，在利用上述神经网络模型之前，需要对其进行训练。也就是说，在本申请的妊娠风险预警管理系统中，还包括用于对所述基于双向门控循环单元的序列编码器、所述基于门控响应的特征动态交互融合模块、所述基于嵌入层的多模态特征融合器和所述基于分类器的预警管理结果生成器进行训练的训练模块。It should be understood that before using the above neural network model, it needs to be trained. That is, in the pregnancy risk warning management system of the present application, it also includes a training module for training the sequence encoder based on the bidirectional gated recurrent unit, the feature dynamic interactive fusion module based on the gated response, the multimodal feature fuser based on the embedding layer, and the warning management result generator based on the classifier.
图6为根据本申请实施例的妊娠风险预警管理系统中训练模块的框图。如图6所示，所述训练模块200，包括：训练数据获取单元210，用于获取训练数据，所述训练数据包括由可穿戴设备采集的被监控孕妇对象的训练生理状态参数的时间序列，其中，所述训练生理状态参数包括训练心率值、训练血压值、训练脉搏频率值以及所述被监控孕妇对象的生理状态是否存在异常的真实值；训练参数样本维度排列单元220，用于将所述训练生理状态参数的时间序列按照参数样本维度进行排列以得到训练心率值的时间序列、训练血压值的时间序列和训练脉搏频率值的时间序列；训练参数时序编码单元230，用于将所述训练心率值的时间序列、所述训练血压值的时间序列和所述训练脉搏频率值的时间序列输入所述基于双向门控循环单元的序列编码器以得到训练心率值时序关联隐含特征向量、训练血压值时序关联隐含特征向量和训练脉搏频率值时序关联隐含特征向量；训练参数时序排列单元240，用于将所述训练心率值的时间序列、所述训练血压值的时间序列和所述训练脉搏频率值的时间序列分别按照时间维度排列为训练心率值时序输入向量、训练血压值时序输入向量和训练脉搏频率值时序输入向量；训练参数多尺度时序特征融合单元250，用于将所述训练心率值时序关联隐含特征向量和所述训练心率值时序输入向量、所述训练血压值时序关联隐含特征向量和所述训练血压值时序输入向量以及所述训练脉搏频率值时序关联隐含特征向量和所述训练脉搏频率值时序输入向量输入所述基于门控响应的特征动态交互融合模块以得到训练心率多尺度门控融合特征向量、训练血压多尺度门控融合特征向量和训练脉搏频率多尺度门控融合特征向量；训练参数多模态特征融合单元260，用于将所述训练血压多尺度门控融合特征向量、所述训练心率多尺度门控融合特征向量和所述训练脉搏频率多尺度门控融合特征向量输入所述基于嵌入层的多模态特征融合器以得到训练生理状态多模态融合特征向量；分类损失单元270，用于将所述训练生理状态多模态融合特征向量输入所述基于分类器的预警管理结果生成器以得到分类损失函数值；预定损失函数值计算单元280，用于计算所述训练生理状态多模态融合特征向量的预定损失函数值；模型训练单元290，用于以所述分类损失函数值和所述预定损失函数值的加权和作为损失函数值，来对所述基于双向门控循环单元的序列编码器、所述基于门控响应的特征动态交互融合模块、所述基于嵌入层的多模态特征融合器和所述基于分类器的预警管理结果生成器进行训练。FIG6 is a block diagram of a training module in a pregnancy risk warning management system according to an embodiment of the present application. As shown in FIG6 , the training module 200 includes: a training data acquisition unit 210 for acquiring training data, wherein the training data includes a time series of training physiological state parameters of the monitored pregnant woman collected by a wearable device, wherein the training physiological state parameters include training heart rate values, training blood pressure values, training pulse frequency values, and a true value of whether the physiological state of the monitored pregnant woman is abnormal; a training parameter sample dimension arrangement unit 220 for arranging the time series of the training physiological state parameters according to the parameter sample dimension to obtain a time series of training heart rate values, a time series of training blood pressure values, and a time series of training pulse frequency values; a training parameter timing encoding unit 230 for encoding the training heart rate values into a time series of training blood pressure values; and a training parameter timing encoding unit 230 for encoding the training heart rate values into a time series of training blood pressure values. The time series of the training blood pressure values and the time series of the training pulse frequency values are input into the sequence encoder based on the bidirectional gated cyclic unit to obtain the training heart rate value time series associated implicit feature vector, the training blood pressure value time series associated implicit feature vector and the training pulse frequency value time series associated implicit feature vector; the training parameter time series arrangement unit 240 is used to arrange the time series of the training heart rate values, the time series of the training blood pressure values and the time series of the training pulse frequency values according to the time dimension into the training heart rate value time series input vector, the training blood pressure value time series input vector and the training pulse frequency value time series input vector respectively; the training parameter multi-scale time series feature fusion unit 250 is used to arrange the training heart rate value time series associated implicit feature vector The training heart rate value and the training blood pressure value time series input vector, the training blood pressure value time series associated implicit feature vector and the training blood pressure value time series input vector, and the training pulse frequency value time series associated implicit feature vector and the training pulse frequency value time series input vector are input into the feature dynamic interactive fusion module based on gated response to obtain the training heart rate multi-scale gated fusion feature vector, the training blood pressure multi-scale gated fusion feature vector and the training pulse frequency multi-scale gated fusion feature vector; the training parameter multimodal feature fusion unit 260 is used to input the training blood pressure multi-scale gated fusion feature vector, the training heart rate multi-scale gated fusion feature vector and the training pulse frequency multi-scale gated fusion feature vector into the multimodal feature fusion module based on embedding layer. a classifier 270 for inputting the training physiological state multimodal fusion feature vector into the classifier-based early warning management result generator to obtain a classification loss function value; a predetermined loss function value calculation unit 280 for calculating a predetermined loss function value of the training physiological state multimodal fusion feature vector; a model training unit 290 for training the sequence encoder based on the bidirectional gated recurrent unit, the feature dynamic interaction fusion module based on the gated response, the multimodal feature fuser based on the embedding layer, and the classifier-based early warning management result generator using the weighted sum of the classification loss function value and the predetermined loss function value as the loss function value.
这里，考虑到所述训练血压多尺度门控融合特征向量、所述训练心率多尺度门控融合特征向量和所述训练脉搏频率多尺度门控融合特征向量分别融合了血压值、心率值和脉搏频率值的源时序分布特征和双向门控循环时序编码特征，将所述训练血压多尺度门控融合特征向量、所述训练心率多尺度门控融合特征向量和所述训练脉搏频率多尺度门控融合特征向量输入基于嵌入层的多模态特征融合器时，得到的所述训练生理状态多模态融合特征向量也会基于相异源样本数据时序分布的多模态特征融合表示而具有基于复杂特征模态表达模式的类回归识别困难，从而影响分类训练效率。Here, considering that the training blood pressure multi-scale gated fusion feature vector, the training heart rate multi-scale gated fusion feature vector and the training pulse frequency multi-scale gated fusion feature vector respectively fuse the source time series distribution characteristics and bidirectional gated cyclic time series coding characteristics of blood pressure values, heart rate values and pulse frequency values, when the training blood pressure multi-scale gated fusion feature vector, the training heart rate multi-scale gated fusion feature vector and the training pulse frequency multi-scale gated fusion feature vector are input into the multimodal feature fuser based on the embedding layer, the obtained training physiological state multimodal fusion feature vector will also have class regression recognition difficulties based on the multimodal feature fusion representation of the time series distribution of different source sample data, thereby affecting the classification training efficiency.
因此，在本申请的技术方案中，在模型训练过程中，进一步引入分类损失函数以外的预定损失函数，也就是，基于损失函数值通过梯度反向传播来训练模型包括以下步骤：计算基于所述训练生理状态多模态融合特征向量的第一生理状态多模态融合权重矩阵和第二生理状态多模态融合权重矩阵，其中，所述第一生理状态多模态融合权重矩阵和第二生理状态多模态融合权重矩阵的第位置的特征值分别是所述生理状态多模态融合特征向量的第特征值和第特征值的均值和差绝对值的二分之一；将所述训练生理状态多模态融合特征向量分别与所述第一生理状态多模态融合权重矩阵和所述第二生理状态多模态融合权重矩阵进行查询式矩阵相乘以得到第一生理状态多模态融合中间向量和第二生理状态多模态融合中间向量；计算所述第一生理状态多模态融合中间向量和所述第二生理状态多模态融合中间向量的向量内积以得到第一生理状态多模态融合损失项；将所述第一生理状态多模态融合权重矩阵和所述第二生理状态多模态融合权重矩阵进行矩阵相乘，并计算结果矩阵的范数以得到第二生理状态多模态融合损失项；将所述第一生理状态多模态融合损失项减去预定权重超参数与所述第二生理状态多模态融合损失项的乘积以得到所述预定损失函数值；基于所述预定损失函数值与所述分类损失函数值的加权和，通过梯度反向传播来优化模型参数。Therefore, in the technical solution of the present application, in the process of model training, a predetermined loss function other than the classification loss function is further introduced, that is, training the model through gradient back propagation based on the loss function value includes the following steps: calculating a first physiological state multimodal fusion weight matrix and a second physiological state multimodal fusion weight matrix based on the training physiological state multimodal fusion feature vector, wherein the first physiological state multimodal fusion weight matrix and the second physiological state multimodal fusion weight matrix are The eigenvalues of the positions are the first Eigenvalues and The mean and half of the absolute value of the difference of the eigenvalues; perform query matrix multiplication of the training physiological state multimodal fusion feature vector with the first physiological state multimodal fusion weight matrix and the second physiological state multimodal fusion weight matrix respectively to obtain the first physiological state multimodal fusion intermediate vector and the second physiological state multimodal fusion intermediate vector; calculate the vector inner product of the first physiological state multimodal fusion intermediate vector and the second physiological state multimodal fusion intermediate vector to obtain the first physiological state multimodal fusion loss term; perform matrix multiplication of the first physiological state multimodal fusion weight matrix and the second physiological state multimodal fusion weight matrix, and calculate the result matrix norm to obtain a second physiological state multimodal fusion loss term; subtract the product of a predetermined weight hyperparameter and the second physiological state multimodal fusion loss term from the first physiological state multimodal fusion loss term to obtain the predetermined loss function value; based on the weighted sum of the predetermined loss function value and the classification loss function value, optimize the model parameters through gradient back propagation.
也就是，以如下生理状态多模态融合损失函数计算所述训练生理状态多模态融合特征向量的预定损失函数值，其中，所述生理状态多模态融合损失函数具体表示为：That is, the predetermined loss function value of the training physiological state multimodal fusion feature vector is calculated using the following physiological state multimodal fusion loss function, wherein the physiological state multimodal fusion loss function is specifically expressed as:
； ;
； ;
； ;
其中，表示所述训练生理状态多模态融合特征向量，表示所述训练生理状态多模态融合特征向量的第个特征值，表示所述训练生理状态多模态融合特征向量的第个特征值，表示所述训练生理状态多模态融合特征向量的第个特征值和第个特征值的均值，是第一心率多尺度时序交互融合权重矩阵，表示所述训练生理状态多模态融合特征向量的第个特征值和第个特征值的差绝对值的二分之一，是第二心率多尺度时序交互融合权重矩阵，表示矩阵乘法运算，为预定权重超参数，表示矩阵的范数，表示所述预定损失函数值。in, represents the multimodal fusion feature vector of the training physiological state, The first multimodal fusion feature vector representing the training physiological state The characteristic values, The first multimodal fusion feature vector representing the training physiological state The characteristic values, The first multimodal fusion feature vector representing the training physiological state The eigenvalues and The mean of the eigenvalues, is the first heart rate multi-scale temporal interactive fusion weight matrix, The first multimodal fusion feature vector representing the training physiological state The eigenvalues and One-half the absolute value of the difference between the eigenvalues, is the second heart rate multi-scale temporal interactive fusion weight matrix, represents the matrix multiplication operation, is the predetermined weight hyperparameter, Represents the matrix norm, represents the predetermined loss function value.
也就是，所述训练生理状态多模态融合损失函数通过所述训练生理状态多模态融合特征向量的近程-远程跨尺度细节链接的结构化特征表示，来进行所述训练生理状态多模态融合特征向量内的细节内积空间的查询式组成，以近似所述训练生理状态多模态融合特征向量的结构化细节交互所提供的链接细节组成的低秩化独立可观测组成，这样，通过以所述训练生理状态多模态融合损失函数来进行训练，就可以通过所述训练生理状态多模态融合特征向量的分布式细节组来在细节复杂性基础上进行细节组分解，以促进所述训练生理状态多模态融合特征向量的复杂特征结构的类回归分解识别，提升分类训练效率。That is, the training physiological state multimodal fusion loss function uses the structured feature representation of the short-range and long-range cross-scale detail links of the training physiological state multimodal fusion feature vector to perform the query composition of the detail inner product space within the training physiological state multimodal fusion feature vector, so as to approximate the low-rank independent observable composition of the link detail composition provided by the structured detail interaction of the training physiological state multimodal fusion feature vector. In this way, by training with the training physiological state multimodal fusion loss function, the distributed detail groups of the training physiological state multimodal fusion feature vector can be used to decompose the detail groups based on the detail complexity, so as to promote the class regression decomposition and recognition of the complex feature structure of the training physiological state multimodal fusion feature vector and improve the classification training efficiency.
综上，根据本申请实施例的妊娠风险预警管理系统被阐明，其采用基于深度学习的人工智能技术对孕妇的生理状态进行实时监测和数据分析，以捕捉到孕妇的心率、血压和脉搏频率的时序变化特性，进而基于孕妇各项生理状态参数的时序多模态融合特征来智能判断孕妇的生理状态是否存在异常。这样，减轻医疗人员的工作负担，保障母婴的健康安全。In summary, the pregnancy risk warning management system according to the embodiment of the present application is explained, which uses artificial intelligence technology based on deep learning to perform real-time monitoring and data analysis on the physiological state of pregnant women, so as to capture the time series change characteristics of the heart rate, blood pressure and pulse frequency of pregnant women, and then intelligently judge whether the physiological state of pregnant women is abnormal based on the time series multimodal fusion characteristics of various physiological state parameters of pregnant women. In this way, the workload of medical personnel is reduced and the health and safety of mothers and babies are guaranteed.
以上结合具体实施例描述了本发明的基本原理，但是，需要指出的是，在本发明中提及的优点、优势、效果等仅是示例而非限制，不能认为这些优点、优势、效果等是本发明的各个实施例必须具备的。另外，上述实施例的具体细节仅是为了示例的作用和便于理解的作用，而非限制，上述细节并不限制本发明为必须采用上述具体的细节来实现。The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述或记载的部分，可以参见其它实施例的相关描述。在本发明所提供的几个实施例中，应该理解到，所揭露的系统和方法，可以通过其它的方式实现。例如，以上所描述的系统实施例仅仅是示意性的，例如，所述单元划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式。所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.
对于本领域技术人员而言，显然本发明不限于上述示范性实施例的细节，而且在不背离本发明的精神或基本特征的情况下，能够以其他的具体形式实现本发明。因此，无论从哪一点来看，均应将实施例看作是示范性的，而且是非限制性的，本发明的范围由所附权利要求而不是上述说明限定，因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本发明内。不应将权利要求中的任何附关联图标记视为限制所涉及的权利要求。It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and range of equivalent elements of the claims be included in the invention. Any reference to a figure in a claim should not be considered as limiting the claim to which it relates.
最后应说明的是，为了例示和描述的目的已经给出了以上描述。此外，以上实施例仅用以说明本发明的技术方案而非限制，尽管参照较佳实施例对本发明进行了详细说明，本领域的普通技术人员应当理解，可以对本发明的技术方案进行修改或等同替换，而不脱离本发明技术方案的精神和范围。Finally, it should be noted that the above description has been given for the purpose of illustration and description. In addition, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411001425.4ACN118553424B (en) | 2024-07-25 | 2024-07-25 | Pregnancy risk early warning management system |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411001425.4ACN118553424B (en) | 2024-07-25 | 2024-07-25 | Pregnancy risk early warning management system |
| Publication Number | Publication Date |
|---|---|
| CN118553424Atrue CN118553424A (en) | 2024-08-27 |
| CN118553424B CN118553424B (en) | 2024-10-25 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411001425.4AActiveCN118553424B (en) | 2024-07-25 | 2024-07-25 | Pregnancy risk early warning management system |
| Country | Link |
|---|---|
| CN (1) | CN118553424B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119094246A (en)* | 2024-11-06 | 2024-12-06 | 福建中信网安信息科技有限公司 | Network security protection operation and maintenance management system and method |
| CN119541874A (en)* | 2025-01-23 | 2025-02-28 | 吉林大学 | Intelligent nursing system and method for breast cancer patients after surgery |
| CN120412987A (en)* | 2025-07-07 | 2025-08-01 | 上海交通大学医学院附属仁济医院 | An AI-assisted diagnosis system for malignant tumors based on an Internet hospital |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114159039A (en)* | 2021-11-12 | 2022-03-11 | 广州三瑞医疗器械有限公司 | An intelligent prenatal fetal heart monitoring model |
| CN117457209A (en)* | 2023-11-08 | 2024-01-26 | 中南大学 | Multi-mode clinical risk prediction method and system based on sampling time interval |
| CN117524451A (en)* | 2023-10-13 | 2024-02-06 | 武汉大学人民医院（湖北省人民医院） | Pregnancy hypertension nursing system based on remote monitoring technology |
| CN118098599A (en)* | 2024-04-22 | 2024-05-28 | 吉林大学 | Pregnancy monitoring management system and method for high-risk pregnant and lying-in women |
| CN118173270A (en)* | 2024-05-14 | 2024-06-11 | 吉林大学 | System and method for assessing risk of postoperative infection in patients |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114159039A (en)* | 2021-11-12 | 2022-03-11 | 广州三瑞医疗器械有限公司 | An intelligent prenatal fetal heart monitoring model |
| CN117524451A (en)* | 2023-10-13 | 2024-02-06 | 武汉大学人民医院（湖北省人民医院） | Pregnancy hypertension nursing system based on remote monitoring technology |
| CN117457209A (en)* | 2023-11-08 | 2024-01-26 | 中南大学 | Multi-mode clinical risk prediction method and system based on sampling time interval |
| CN118098599A (en)* | 2024-04-22 | 2024-05-28 | 吉林大学 | Pregnancy monitoring management system and method for high-risk pregnant and lying-in women |
| CN118173270A (en)* | 2024-05-14 | 2024-06-11 | 吉林大学 | System and method for assessing risk of postoperative infection in patients |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119094246A (en)* | 2024-11-06 | 2024-12-06 | 福建中信网安信息科技有限公司 | Network security protection operation and maintenance management system and method |
| CN119541874A (en)* | 2025-01-23 | 2025-02-28 | 吉林大学 | Intelligent nursing system and method for breast cancer patients after surgery |
| CN120412987A (en)* | 2025-07-07 | 2025-08-01 | 上海交通大学医学院附属仁济医院 | An AI-assisted diagnosis system for malignant tumors based on an Internet hospital |
| Publication number | Publication date |
|---|---|
| CN118553424B (en) | 2024-10-25 |
| Publication | Publication Date | Title |
|---|---|---|
| CN118553424B (en) | Pregnancy risk early warning management system | |
| Xia et al. | Generative adversarial network with transformer generator for boosting ECG classification | |
| US20200337580A1 (en) | Time series data learning and analysis method using artificial intelligence | |
| CN111160139A (en) | ECG signal processing method, device and terminal equipment | |
| JP2018068752A (en) | Machine learning device, machine learning method and program | |
| Imran Molla et al. | Cardiotocogram data classification using random forest based machine learning algorithm | |
| Xiao et al. | A deep feature fusion network for fetal state assessment | |
| CN117457209A (en) | Multi-mode clinical risk prediction method and system based on sampling time interval | |
| Buettner et al. | Heartbeat classification with spiking neural networks on the loihi neuromorphic processor | |
| Zhang et al. | Wearable sensors based automatic box and block test system | |
| CN116631626A (en) | A patient clinical risk assessment method, device, equipment and medium | |
| Mineo et al. | A convolutional-transformer model for ffr and ifr assessment from coronary angiography | |
| CN118806270A (en) | Nursing monitoring system and method based on anti-bedsore intelligent mattress | |
| CN113855008A (en) | Wearable body fluid monitoring device and operation method thereof | |
| Singh et al. | A Comparative Study of Machine Learning and Deep Learning Methods for Detecting Thyroid Disease: An Experimental Investigation | |
| Mondal et al. | Deep learning-based diabetic retinopathy detection for multiclass imbalanced data | |
| Jaganathan et al. | Modelling an effectual feature selection approach for predicting down syndrome using machine learning approaches | |
| Purwono et al. | Analysis of the relationship between fetal health prediction features with machine learning feyn qlattice regression model | |
| Weihua et al. | Multi-feature Fusion Recognition Method of Heart Sound Signal Based on GADF and CNN-ViT | |
| Nwokoro et al. | Unlocking maternal outcome prediction potential: a comprehensive analysis of the convXGB model integrating XGBoost and deep learning | |
| Nguyen-Trong et al. | Graph Convolutional Network for Occupational Disease Prediction with Multiple Dimensional Data | |
| Abinaya et al. | Enhanced Fetal Health Prediction using Deep Learning and Ensemble Methods | |
| Zha et al. | A Fuzzy Radial Basis Inference Network With Grouped Signal Feature Embedding and Its Application in Multi-Source Signals Classification | |
| Ramachandran et al. | from Cardiotocographic Data | |
| Subburaj et al. | Prediction of Fetal Brain Anomalies Using Deep Learning Models |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |