Postpartum hemorrhage risk assessment system based on medical information processing

Abstract

The present invention discloses a postpartum hemorrhage risk assessment system based on medical information processing, including a physiological data acquisition module, a data optimization module, a postpartum hemorrhage risk assessment model establishment module and a postpartum hemorrhage risk assessment module. The present invention belongs to the field of medical information processing technology, and specifically refers to a postpartum hemorrhage risk assessment system based on medical information processing. The scheme captures the key change trend of maternal physiological data through derivative features, improves the recognition ability of high-risk sequences; realizes personalized matching based on dynamic programming, and enhances the screening efficiency of high-risk samples; improves the utilization rate of maternal physiological sample data based on length correction and fusion sorting; thereby ensuring the accuracy of subsequent postpartum hemorrhage risk assessment results; processes physiological time series and maternal background information separately, and introduces a hierarchical attention mechanism to adapt to risk sources in different scenarios; provides dynamic risk assessment results based on burst adjustment weights and regional enhancement weights; thereby improving the accuracy of the final risk assessment results.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information processing, and in particular to a postpartum hemorrhage risk assessment system based on medical information processing. Background Art

[0002] The postpartum hemorrhage risk assessment system is a comprehensive tool based on medical data analysis, machine learning or rule models. It aims to predict, evaluate and warn the risk of postpartum hemorrhage by collecting and analyzing the multi-dimensional data of parturients. However, the general postpartum hemorrhage risk assessment system does not fully consider the differences in the physiological time series rhythm and individual characteristics of parturients, resulting in a low proportion of effectively used maternal physiological data, which in turn leads to poor accuracy in the final postpartum hemorrhage risk assessment; the general postpartum hemorrhage risk assessment system cannot adapt to the risk sources in different scenarios, and only relies on dynamic physiological indicators while ignoring the significant impact of background information on risks, which leads to poor results in bleeding risk assessment. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a postpartum hemorrhage risk assessment system based on medical information processing. In view of the fact that the general postpartum hemorrhage risk assessment system does not fully consider the physiological time series rhythm differences and individual characteristics of the parturient, resulting in a low proportion of effectively utilized maternal physiological data, and thus resulting in poor accuracy of the final postpartum hemorrhage risk assessment, this solution captures the key change trends of maternal physiological data through derivative features to improve the recognition ability of high-risk sequences; realizes personalized matching based on dynamic programming to enhance the screening efficiency of high-risk samples; improves the utilization rate of maternal physiological sample data based on length correction and fusion sorting; thereby ensuring the accuracy of subsequent postpartum hemorrhage risk assessment results; in view of the fact that the general postpartum hemorrhage risk assessment system cannot adapt to the risk sources in different scenarios, only relies on dynamic physiological indicators and ignores the significant impact of background information on risks, thereby resulting in poor bleeding risk assessment results, this solution processes the physiological time series and maternal background information separately, and introduces a hierarchical attention mechanism to fuse physiological indicators, background information and historical risk values ​​to adapt to the risk sources in different scenarios; provides dynamic risk assessment results based on burst adjustment weights and regional enhancement weights; thereby improving the accuracy of the final risk assessment results.

[0004] The technical solution adopted by the present invention is as follows: the postpartum hemorrhage risk assessment system based on medical information processing provided by the present invention includes a physiological data acquisition module, a data optimization module, a postpartum hemorrhage risk assessment model establishment module and a postpartum hemorrhage risk assessment module;

[0005] The physiological data collection module collects historical postpartum hemorrhage medical data;

[0006] The data optimization module selects the historical physiological time series data that best matches the standard high-risk pattern through dynamic programming and trend feature matching technology, and optimizes the model training data set;

[0007] The postpartum hemorrhage risk assessment model establishment module is based on a dual encoder architecture and a multi-channel attention mechanism, which comprehensively processes the physiological time series and background information of the parturient, dynamically predicts the risk of postpartum hemorrhage and provides multi-step risk assessment results;

[0008] The postpartum hemorrhage risk assessment module implements postpartum hemorrhage risk assessment based on the established postpartum hemorrhage risk assessment model.

[0009] Furthermore, in the physiological data collection module, the historical postpartum hemorrhage medical data includes the physiological time series data and delivery outcomes of the parturient; the delivery outcomes are used as data labels; and the data are cleaned, converted and standardized.

[0010] Furthermore, the data optimization module is to match the historical physiological time series Compared with the standard high-risk time series Q, the optimal data set is screened; specifically including the following:

[0011] Calculate the similarity between the standard high-risk time series and each historical physiological time series, and capture the changing trend of physiological indicators through derivative values; expressed as: ; ; where d ij is the i-th point in the standard sequence and the jth point in the history sequence The local distance of ; D(·) is the derivative eigenvalue; and They are the i-1th point and i+1th point in the standard sequence respectively; is the trend parameter;

[0012] The cumulative distance is dynamically programmed to adapt to the rhythm changes of the maternal physiological time series, and a distortion penalty term is introduced, which is expressed as: ; ; Where DL(·) is the cumulative distance; i and j are the point indices of the sequence; is the distortion penalty coefficient; is the distortion penalty term;

[0013] Based on the dynamic programming cumulative distance, the optimal path is tracked backward to obtain the optimal total alignment distance, which is expressed as: ; Where DD(·) is the total distance of the best alignment between the standard high-risk time series and the historical physiological time series; K1 is the total number of path points; is the local distance of the k1th point on the optimal path; is the distortion penalty of the optimal path point;

[0014] Correct the length difference between the historical sample sequence and the standard sequence, expressed as: ; Wherein, DSL is sequence length correction; is the sequence length difference; is the difference in the starting values ​​of the sequences; is the difference in the sequence endpoint values; and is the correction weight coefficient;

[0015] Fusion sorting: A similarity threshold is set in advance, and the top K historical samples that best match the standard high-risk samples are selected as the input data set; expressed as: ;in, The samples are selected based on the sorting results; is the best alignment total distance sorting, selecting the first L1 samples; It is to correct the sorting according to the sequence length and select the first L2 samples; It is a threshold filter that removes samples below the similarity threshold.

[0016] Furthermore, the postpartum hemorrhage risk assessment model establishment module takes the optimized historical postpartum hemorrhage medical data as input, and specifically includes the following contents:

[0017] Encoder design; The model for postpartum hemorrhage risk assessment adopts a dual encoder architecture to process the dynamic physiological characteristics and background information of the parturient respectively; the physiological indicator encoder extracts the local features of the physiological time series through extended time convolution and captures the dependencies between features through multi-channel attention; it is expressed as: ; ; ; The background information encoder embeds and maps the maternal background information to a high-dimensional space; it is expressed as: ; Where F(·) is the local feature extracted by the extended temporal convolution, t is the time step; f(·) is the weight of the convolution filter; d(t) is the temporal expansion factor; is the regulating factor; is a smoothing term; k is the convolution kernel size, i1 is the convolution kernel size index; is time step t forward The characteristic value of i1 represents the value of the physiological index sampled from the historical time point; is multi-channel attention; Q, K, and V are query vector, key vector, and value vector respectively; is the scaling factor; dmax is the maximum expansion factor; min(·) is the minimum value; softmax(·) is the softmax function; It is the embedded background information representation; is the background information of the parturient; Embedding(·) is to embed the background information into a high-dimensional space; is the feature importance weight;

[0018] Decoder design: The decoder uses the physiological time series features and maternal background information extracted by the encoder to predict the risk of postpartum hemorrhage in the future time step through multi-channel attention; and introduces a hierarchical attention mechanism; expressed as: ;in, is the output of the decoder, which represents the predicted value of postpartum hemorrhage risk in the future time step; , and They are the output of the physiological indicator encoder, the output of the background information encoder and the input of the decoder; the output of the physiological indicator encoder represents the feature representation of the physiological time series; the output of the background information encoder represents the background information representation of the parturient; the input of the decoder represents the known risk value of the historical time step; , and is the attention weight;

[0019] Multi-step prediction: Output the risk assessment value of postpartum hemorrhage in the next O time steps through the decoder; expressed as: ; Introduce burst adjustment weight , dynamically adjust based on the risk value growth rate; introduce feature contribution weights , weighting the impact of risks according to key physiological indicators; introducing regional enhancement weights , locally zoom in on the postpartum hemorrhage assessment at the early time step; the final output is obtained, expressed as: ; ; ; ; ; ;in, is the risk assessment value of postpartum hemorrhage in the future O time steps; , and are the risk assessment values ​​of postpartum hemorrhage in the future 1 time step, 2 time steps and O time steps respectively; R is the final output risk assessment value of postpartum hemorrhage; is the weighting factor of time step t; I is the current time step; γ is the adjustment parameter; and are the risk prediction values ​​at time step t and time step t-1 respectively; is a smoothing term; n is the number of risk features, k is the risk feature index; is the characteristic change rate; is the feature importance weight; is the critical time step; is the width parameter; and are the risk characteristic values ​​at time step t and time step t-1 respectively;

[0020] Loss function design; based on cross entropy loss And introduce dynamic change penalty to design the loss function of the postpartum hemorrhage risk assessment model , expressed as: ;in, is the loss weight; T is the total time step; It is the actual trend change;

[0021] Model determination; pre-set the maximum number of training times and prediction threshold; use the optimized historical postpartum hemorrhage medical data as a data set and divide it into a training set and a test set; when the loss of the postpartum hemorrhage risk assessment model for the training set converges or reaches the maximum number of iterations, the training of the postpartum hemorrhage risk assessment model is completed; when the prediction accuracy of the trained postpartum hemorrhage risk assessment model for the test set is higher than the prediction threshold, the postpartum hemorrhage risk assessment model is established, otherwise the model parameters are optimized based on the gray wolf search algorithm and the data set is re-divided to train the model.

[0022] Furthermore, the postpartum hemorrhage risk assessment module is based on the established postpartum hemorrhage risk assessment model, collects the physiological time series data of the parturient in real time and inputs it into the postpartum hemorrhage risk assessment model, and uses the delivery outcome output by the postpartum hemorrhage risk assessment model as the risk assessment result.

[0023] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0024] (1) The general postpartum hemorrhage risk assessment system does not fully consider the differences in the physiological time series rhythm and individual characteristics of the parturient, resulting in a low proportion of effectively utilized maternal physiological data, which in turn leads to poor accuracy in the final postpartum hemorrhage risk assessment. This solution captures the key change trends of maternal physiological data through derivative features to improve the ability to identify high-risk sequences; achieves personalized matching based on dynamic programming to enhance the screening efficiency of high-risk samples; improves the utilization rate of maternal physiological sample data based on length correction and fusion sorting; and thus ensures the accuracy of subsequent postpartum hemorrhage risk assessment results.

[0025] (2) In view of the problem that the general postpartum hemorrhage risk assessment system cannot adapt to the risk sources in different scenarios, and only relies on dynamic physiological indicators while ignoring the significant impact of background information on risk, which leads to poor bleeding risk assessment results, this scheme processes the physiological time series and maternal background information separately, and introduces a hierarchical attention mechanism to integrate physiological indicators, background information and historical risk values ​​to adapt to the risk sources in different scenarios; it provides dynamic risk assessment results based on sudden adjustment weights and regional enhancement weights; and thus improves the accuracy of the final risk assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A schematic diagram of the process of the postpartum hemorrhage risk assessment system based on medical information processing provided by the present invention;

[0027] Figure 2 Schematic diagram of the process for establishing modules for the postpartum hemorrhage risk assessment model.

[0028] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0030] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0031] Example 1, see Figure 1 , the postpartum hemorrhage risk assessment system based on medical information processing provided by the present invention includes a physiological data acquisition module, a data optimization module, a postpartum hemorrhage risk assessment model establishment module and a postpartum hemorrhage risk assessment module;

[0032] The physiological data collection module collects historical postpartum hemorrhage medical data; and sends the data to the data optimization module;

[0033] The data optimization module receives the data sent by the physiological data acquisition module; through dynamic programming and trend feature matching technology, it screens the historical physiological time series data that best matches the standard high-risk model, optimizes the model training data set; and sends the data to the postpartum hemorrhage risk assessment model establishment module;

[0034] The postpartum hemorrhage risk assessment model establishment module receives data sent by the data optimization module; based on the dual encoder architecture and multi-channel attention mechanism, comprehensively processes the physiological time series and background information of the parturient, dynamically predicts the risk of postpartum hemorrhage and provides multi-step risk assessment results; and sends the data to the postpartum hemorrhage risk assessment module;

[0035] The postpartum hemorrhage risk assessment module receives data sent by the postpartum hemorrhage risk assessment model establishment module; and implements postpartum hemorrhage risk assessment based on the established postpartum hemorrhage risk assessment model.

[0036] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the physiological data acquisition module, the historical postpartum hemorrhage medical data includes the physiological time series data and delivery outcomes of the parturient; the delivery outcomes are used as data labels; and the data is cleaned, converted and standardized; the delivery outcomes include normal, mild bleeding, moderate bleeding and severe bleeding; the physiological time series data of the parturient includes hemodynamic indicators, coagulation function-related indicators, blood volume indicators, uterine status indicators, respiratory function indicators, time, bleeding volume, speed and background information; the background information includes age, weight, height, occupation, lifestyle, previous delivery history, uterine surgery history, previous postpartum hemorrhage history, pregnancy information, disease information and drug use information.

[0037] Example 3, see Figure 1 This embodiment is based on the above embodiment, and the data optimization module is to match the historical physiological time series Compared with the standard high-risk time series Q, the optimal data set is screened; specifically including the following:

[0038] Calculate the similarity between the standard high-risk time series and each historical physiological time series, and capture the changing trend of physiological indicators through derivative values; expressed as: ; ; where d ij is the i-th point in the standard sequence and the jth point in the history sequence The local distance of ; D(·) is the derivative eigenvalue; and They are the i-1th point and i+1th point in the standard sequence respectively; is the trend parameter; The value range is [0.1,1.0];

[0039] The cumulative distance is dynamically programmed to adapt to the rhythm changes of the maternal physiological time series, and a distortion penalty term is introduced, which is expressed as: ; ; Where DL(·) is the cumulative distance; i and j are the point indices of the sequence; is the distortion penalty coefficient; is the distortion penalty term; The value range is [0.01,0.5];

[0040] Based on the dynamic programming cumulative distance, the optimal path is tracked backward to obtain the optimal total alignment distance, which is expressed as: ; Where DD(·) is the total distance of the best alignment between the standard high-risk time series and the historical physiological time series; K1 is the total number of path points; is the local distance of the k1th point on the optimal path; is the distortion penalty of the optimal path point;

[0041] Correct the length difference between the historical sample sequence and the standard sequence, expressed as: ; Wherein, DSL is sequence length correction; is the sequence length difference; is the difference in the starting values ​​of the sequences; is the difference in the sequence endpoint values; and is the correction weight coefficient; The value range is [0.1,1.0]; The value range is [0.1,1.0];

[0042] Fusion sorting: A similarity threshold is set in advance, and the top K historical samples that best match the standard high-risk samples are selected as the input data set; expressed as: ;in, The samples are selected based on the sorting results; is the best alignment total distance sorting, selecting the first L1 samples; It is to correct the sorting according to the sequence length and select the first L2 samples; It is a threshold filter that removes samples below the similarity threshold; the similarity threshold value range is [0.7, 1.0].

[0043] By performing the above operations, the general postpartum hemorrhage risk assessment system does not fully consider the differences in the physiological time series rhythm and individual characteristics of the parturient, resulting in a low proportion of effectively utilized maternal physiological data, and then leading to poor accuracy of the final postpartum hemorrhage risk assessment. This solution captures the key change trends of maternal physiological data through derivative features to improve the recognition ability of high-risk sequences; realizes personalized matching based on dynamic programming to enhance the screening efficiency of high-risk samples; improves the utilization rate of maternal physiological sample data based on length correction and fusion sorting; and thus ensures the accuracy of subsequent postpartum hemorrhage risk assessment results.

[0044] Example 4, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The postpartum hemorrhage risk assessment model establishment module takes the optimized historical postpartum hemorrhage medical data as input, and specifically includes the following contents:

[0045] Encoder design; The model for postpartum hemorrhage risk assessment adopts a dual encoder architecture to process the dynamic physiological characteristics and background information of the parturient respectively; the physiological indicator encoder extracts the local features of the physiological time series through extended time convolution and captures the dependencies between features through multi-channel attention; it is expressed as: ; ; ; The background information encoder embeds and maps the maternal background information to a high-dimensional space; it is expressed as: ; Where F(·) is the local feature extracted by the extended temporal convolution, t is the time step; f(·) is the weight of the convolution filter; d(t) is the temporal expansion factor; is the regulating factor; is a smoothing term; k is the convolution kernel size, i1 is the convolution kernel size index; is time step t forward The characteristic value of i1 represents the value of the physiological index sampled from the historical time point; is multi-channel attention; Q, K, and V are query vector, key vector, and value vector respectively; is the scaling factor; dmax is the maximum expansion factor; min(·) is the minimum value; softmax(·) is the softmax function; It is the embedded background information representation; is the background information of the parturient; Embedding(·) is to embed the background information into a high-dimensional space; is the feature importance weight; The value range is [1,10]; the value range of dmax is [5,50]; The value range is [1e-5, 1e-2]; The value range is [0,1];

[0046] Decoder design: The decoder uses the physiological time series features and maternal background information extracted by the encoder to predict the risk of postpartum hemorrhage in the future time step through multi-channel attention; and introduces a hierarchical attention mechanism; expressed as: ;in, is the output of the decoder, which represents the predicted value of postpartum hemorrhage risk in the future time step; , and They are the output of the physiological indicator encoder, the output of the background information encoder and the input of the decoder; the output of the physiological indicator encoder represents the feature representation of the physiological time series; the output of the background information encoder represents the background information representation of the parturient; the input of the decoder represents the known risk value of the historical time step; , and is the attention weight; the attention weight range is [0,1], and the sum is 1;

[0047] Multi-step prediction: Output the risk assessment value of postpartum hemorrhage in the next O time steps through the decoder; expressed as: ; Introduce burst adjustment weight , dynamically adjust based on the risk value growth rate; introduce feature contribution weights , weighting the impact of risks according to key physiological indicators; introducing regional enhancement weights , locally zoom in on the postpartum hemorrhage assessment at the early time step; the final output is obtained, expressed as: ; ; ; ; ; ;in, is the risk assessment value of postpartum hemorrhage in the future O time steps; , and are the risk assessment values ​​of postpartum hemorrhage in the future 1 time step, 2 time steps and O time steps respectively; R is the final output risk assessment value of postpartum hemorrhage; is the weighting factor of time step t; I is the current time step; γ is the adjustment parameter; and are the risk prediction values ​​at time step t and time step t-1 respectively; is a smoothing term; n is the number of risk features, k is the risk feature index; is the characteristic change rate; is the feature importance weight; is the critical time step; is the width parameter; and are the risk characteristic values ​​at time step t and time step t-1 respectively; the value range of γ is [0,1]; The value range is [1,10]; The value range is [1, T];

[0048] Loss function design; based on cross entropy loss And introduce dynamic change penalty to design the loss function of the postpartum hemorrhage risk assessment model , expressed as: ;in, is the loss weight; T is the total time step; It is the actual trend change; The value range is [0.1,10];

[0049] Model determination; pre-set the maximum number of training times and prediction threshold; use the optimized historical postpartum hemorrhage medical data as a data set and divide it into a training set and a test set; when the loss of the postpartum hemorrhage risk assessment model for the training set converges or reaches the maximum number of iterations, the training of the postpartum hemorrhage risk assessment model is completed; when the prediction accuracy of the trained postpartum hemorrhage risk assessment model for the test set is higher than the prediction threshold, the postpartum hemorrhage risk assessment model is established, otherwise the model parameters are optimized based on the gray wolf search algorithm and the data set is re-divided to train the model.

[0050] By performing the above operations, the general postpartum hemorrhage risk assessment system cannot adapt to the risk sources in different scenarios, only relies on dynamic physiological indicators and ignores the significant impact of background information on risk, which leads to poor bleeding risk assessment effect. This solution processes physiological time series and maternal background information separately, and introduces a hierarchical attention mechanism to integrate physiological indicators, background information and historical risk values ​​to adapt to the risk sources in different scenarios; provides dynamic risk assessment results based on sudden adjustment weights and regional enhancement weights; and thus improves the accuracy of the final risk assessment results.

[0051] Example 5, see Figure 1 This embodiment is based on the above embodiment. The postpartum hemorrhage risk assessment module is based on the established postpartum hemorrhage risk assessment model. The physiological time series data of the parturient is collected in real time and input into the postpartum hemorrhage risk assessment model. The delivery outcome output by the postpartum hemorrhage risk assessment model is used as the risk assessment result. When the output is moderate bleeding or severe bleeding, relevant personnel are warned.

[0052] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0053] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0054] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A postpartum hemorrhage risk assessment system based on medical information processing, characterized by: The system includes a physiological data acquisition module, a data optimization module, a postpartum hemorrhage risk assessment model establishment module and a postpartum hemorrhage risk assessment module; The physiological data collection module collects historical postpartum hemorrhage medical data; The data optimization module selects the historical physiological time series data that best matches the standard high-risk pattern through dynamic programming and trend feature matching technology, and optimizes the model training data set; The postpartum hemorrhage risk assessment model building module is based on a dual encoder architecture and a multi-channel attention mechanism, which comprehensively processes the physiological time series and background information of the parturient, dynamically predicts the risk of postpartum hemorrhage and provides multi-step risk assessment results; The postpartum hemorrhage risk assessment module implements postpartum hemorrhage risk assessment based on the established postpartum hemorrhage risk assessment model; The postpartum hemorrhage risk assessment model establishment module includes multi-step prediction; specifically: outputting the postpartum hemorrhage risk assessment value of the next O time steps through the decoder; expressed as: ; Introduce burst adjustment weight , dynamically adjusted based on the risk value growth rate; Introducing feature contribution weights , weighting the impact of risks according to key physiological indicators; introducing regional enhancement weights , locally zoom in on the postpartum hemorrhage assessment at the early time step; the final output is obtained, expressed as: ; ; ; ; ; ;in, is the risk assessment value of postpartum hemorrhage in the future O time steps; , and are the risk assessment values ​​of postpartum hemorrhage in the future 1 time step, 2 time steps and O time steps respectively; R is the final output risk assessment value of postpartum hemorrhage; is the weighting factor of time step t; I is the current time step; γ is the adjustment parameter; and are the risk prediction values ​​at time step t and time step t-1 respectively; is a smoothing term; n is the number of risk features, k is the risk feature index; is the characteristic change rate; is the feature importance weight; is the critical time step; and are the risk characteristic values ​​at time step t and time step t-1 respectively; is the width parameter.

2. The postpartum hemorrhage risk assessment system based on medical information processing according to claim 1, characterized in that: The data optimization module is to match the historical physiological time series Compared with the standard high-risk time series Q, the optimal data set is screened; Includes the following: Calculate the similarity between the standard high-risk time series and each historical physiological time series, and capture the changing trend of physiological indicators through derivative values; expressed as: ; ; where d ij is the i-th point in the standard high-risk time series and the jth point in the historical physiological time series The local distance of ; D(·) is the derivative eigenvalue; and They are the i-1th point and i+1th point of the standard high-risk time series respectively; is the trend parameter; The cumulative distance is dynamically programmed to adapt to the rhythm changes of the maternal physiological time series, and a distortion penalty term is introduced, which is expressed as: ; ; Where DL(·) is the cumulative distance; i and j are the point indices of the sequence; is the distortion penalty coefficient; is the distortion penalty term; Based on the dynamic programming cumulative distance, the optimal path is tracked backward to obtain the optimal total alignment distance, which is expressed as: ; Where DD(·) is the total distance of the best alignment between the standard high-risk time series and the historical physiological time series; K1 is the total number of path points; is the local distance of the k1th point on the optimal path; is the distortion penalty of the optimal path point; The difference in length between the corrected historical physiological time series and the standard high-risk time series is expressed as: ; Wherein, DSL is sequence length correction; is the sequence length difference; is the difference in the starting values ​​of the sequences; is the difference in the sequence endpoint values; and is the correction weight coefficient; Fusion sorting: A similarity threshold is set in advance, and the top K historical samples that best match the standard high-risk samples are selected as the input data set; expressed as: ;in, The samples are selected based on the sorting results; is the best alignment total distance sorting, selecting the first L1 samples; It is to correct the sorting according to the sequence length and select the first L2 samples; It is a threshold filter that removes samples below the similarity threshold.

3. The postpartum hemorrhage risk assessment system based on medical information processing according to claim 2, characterized in that: The postpartum hemorrhage risk assessment model establishment module takes the optimized historical postpartum hemorrhage medical data as input, and specifically includes the following contents: Encoder design; The model for postpartum hemorrhage risk assessment adopts a dual encoder architecture to process maternal dynamic physiological characteristics and background information respectively; The physiological indicator encoder extracts local features of physiological time series through extended temporal convolution and captures the dependencies between features through multi-channel attention; it is expressed as: ; ; ; The background information encoder embeds and maps the maternal background information to a high-dimensional space; it is expressed as: ; Where F(·) is the local feature extracted by the extended temporal convolution, t is the time step; f(·) is the weight of the convolution filter; d(t) is the temporal expansion factor; is the regulating factor; is a smoothing term; k is the convolution kernel size, i1 is the convolution kernel size index; is time step t forward The characteristic value of i1 represents the value of the physiological index sampled from the historical time point; is multi-channel attention; Q, K, and V are query vector, key vector, and value vector respectively; is the scaling factor; dmax is the maximum expansion factor; min(·) is the minimum value; softmax(·) is the softmax function; It is the embedded background information representation; is the background information of the parturient; Embedding(·) is to embed the background information into a high-dimensional space; is the feature importance weight; Decoder design: The decoder uses the physiological time series features and maternal background information extracted by the encoder to predict the risk of postpartum hemorrhage in the future time step through multi-channel attention; and introduces a hierarchical attention mechanism; expressed as: ;in, is the output of the decoder, which represents the predicted value of postpartum hemorrhage risk in the future time step; , and They are the output of the physiological indicator encoder, the output of the background information encoder and the input of the decoder; the output of the physiological indicator encoder represents the feature representation of the physiological time series; the output of the background information encoder represents the background information representation of the parturient; the input of the decoder represents the known risk value of the historical time step; , and is the attention weight; Loss function design; based on cross entropy loss And introduce dynamic change penalty to design the loss function of the postpartum hemorrhage risk assessment model , expressed as: ;in, is the loss weight; T is the total time step; It is the actual trend change; Model determination; pre-set the maximum number of training times and prediction threshold; use the optimized historical postpartum hemorrhage medical data as a data set and divide it into a training set and a test set; when the loss of the postpartum hemorrhage risk assessment model for the training set converges or reaches the maximum number of iterations, the training of the postpartum hemorrhage risk assessment model is completed; when the prediction accuracy of the trained postpartum hemorrhage risk assessment model for the test set is higher than the prediction threshold, the postpartum hemorrhage risk assessment model is established, otherwise the model parameters are optimized based on the gray wolf search algorithm and the data set is re-divided to train the model.

4. The postpartum hemorrhage risk assessment system based on medical information processing according to claim 3 is characterized in that: In the physiological data collection module, the historical postpartum hemorrhage medical data includes the physiological time series data and delivery outcomes of the parturient; the delivery outcomes are used as data labels; and the data are cleaned, converted and standardized.

5. The postpartum hemorrhage risk assessment system based on medical information processing according to claim 4 is characterized in that: The postpartum hemorrhage risk assessment module is based on an established postpartum hemorrhage risk assessment model, collects the physiological time series data of the parturient in real time and inputs it into the postpartum hemorrhage risk assessment model, and uses the delivery outcome output by the postpartum hemorrhage risk assessment model as the risk assessment result.

Images