A vital sign prediction system for pediatric clinical assistance

Abstract

The present invention discloses a vital sign prediction system for pediatric clinical assistance, which belongs to the field of intelligent medical technology and includes a data preparation module, a data preprocessing module, a feature selection module, a vital sign prediction module and a prediction result feedback module. The present invention adopts the dwarf mongoose optimization algorithm combined with spider monkey optimization for feature selection, and by combining the advantages of the two optimization algorithms, improves the calculation speed and removes redundant features, reduces the risk of model overfitting, and improves the performance of the prediction model, thereby enhancing the practicality and effectiveness of the vital sign prediction system as a whole; adopts a time convolution long-term parallel model for vital sign prediction, and sets a time convolution subnet and a long-term memory subnet in parallel, so that the model can efficiently capture local details and global trends at the same time, and is more suitable for capturing short-term fluctuations and long-term laws in vital sign related data, thereby improving the model prediction effect and enhancing the accuracy of the vital sign prediction system as a whole.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent medical technology, and specifically refers to a vital sign prediction system for pediatric clinical assistance. Background Art

[0002] The vital signs prediction system for pediatric clinical assistance is an intelligent system that uses machine learning technology to predict changes in children's vital signs in real time. It is designed to assist pediatricians in clinical decision-making, reduce the workload of medical staff, and improve the efficiency and accuracy of children's health management.

[0003] However, in the existing vital signs prediction process, there are technical problems such as the data dimension involved in vital signs prediction is high and feature redundancy is prone to occur, which leads to model overfitting and inability to effectively predict unknown data, thus affecting the overall practicality and effectiveness of the system; there is a technical problem that vital signs-related data have the characteristics of high variability and cyclical trends, and a single model is difficult to handle these two characteristics at the same time, resulting in poor model prediction results and affecting the overall accuracy of the system. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a vital sign prediction system for pediatric clinical assistance. In view of the technical problems that in the existing vital sign prediction process, the data dimension involved in the vital sign prediction is high and feature redundancy is prone to occur, resulting in model overfitting and inability to effectively predict unknown data, thereby affecting the overall practicality and effectiveness of the system, this solution creatively adopts the dwarf mongoose optimization algorithm combined with spider monkey optimization for feature selection. By combining the advantages of the two optimization algorithms, the calculation speed is improved and redundant features are removed, the risk of model overfitting is reduced, and the performance of the prediction model is improved, thereby enhancing the vital signs. The practicality and effectiveness of the overall vital sign prediction system; in view of the fact that in the existing vital signs prediction process, the vital signs related data have the characteristics of high variability and periodic trends, and a single model is difficult to handle these two characteristics at the same time, resulting in poor model prediction effect and affecting the overall accuracy of the system. This solution creatively adopts the time convolution long-term and short-term parallel model for vital signs prediction, and sets the time convolution subnet and long-term and short-term memory subnet in parallel, so that the model can efficiently capture local details and global trends at the same time, and is more suitable for capturing short-term fluctuations and long-term laws in vital signs related data, thereby improving the model prediction effect and enhancing the overall accuracy of the vital signs prediction system.

[0005] The technical solution adopted by the present invention is as follows: A vital sign prediction system for pediatric clinical assistance provided by the present invention comprises: a data preparation module, a data preprocessing module, a feature selection module, a vital sign prediction module and a prediction result feedback module;

[0006] The data preparation module obtains comprehensive data of the children's ward through data preparation, and sends the comprehensive data of the children's ward to the data preprocessing module;

[0007] The data preprocessing module obtains high-quality data of the children's ward through data preprocessing, and sends the high-quality data of the children's ward to the feature selection module;

[0008] The feature selection module uses the dwarf mongoose optimization algorithm combined with spider monkey optimization to perform feature selection to obtain the optimal features of the children's ward, and sends the optimal features of the children's ward to the vital sign prediction module;

[0009] The vital sign prediction module uses a time convolution long-term and short-term parallel model to predict vital signs, obtain vital sign prediction information, and send the vital sign prediction information to the prediction result feedback module;

[0010] The prediction result feedback module automatically generates a vital sign prediction report through prediction result feedback.

[0011] Furthermore, in the data preparation module, it is used to collect the original data required for vital signs prediction, specifically, by collecting children's vital signs data and ward environment data to obtain comprehensive data of children's wards.

[0012] Furthermore, in the data preprocessing module, the comprehensive data of the children’s ward were preprocessed by deleting duplicate values, interpolating missing values, and standardizing data to obtain high-quality data of the children’s ward.

[0013] Furthermore, in the feature selection module, there are provided an algorithm initialization unit, a leader position update unit, an individual position update unit, a stagnation detection and scout movement unit, an iterative calculation unit, a dwarf mongoose optimization algorithm construction unit combined with spider monkey optimization, and a children's ward optimal feature calculation unit, including the following contents:

[0014] An algorithm initialization unit randomly initializes individual positions in the feature space of high-quality data in the children's ward, sets the number of individuals and the maximum number of iterations, and uses the prediction accuracy of the vital sign prediction model as a fitness function. The individual positions are used to represent feature subsets;

[0015] The leader position update unit updates the local leader position according to the relative fitness value and updates the global leader position according to the fitness value, including the following:

[0016] The fitness function is used to evaluate the fitness value of each individual's current position and calculate the relative fitness value. The calculation formula of the relative fitness value is:

[0017] ;

[0018] In the formula, is the relative fitness value, i is the individual index, used to represent the current individual, fit i is the fitness value of the i-th individual, n is the number of individuals in the current individual's domain;

[0019] Update the local leader position, specifically, in each individual area, select the individual position with the highest relative fitness value as the local leader position;

[0020] Update the global leader position, specifically, select the individual position with the highest fitness value among all individuals as the global leader position;

[0021] The individual position update unit uses the spider monkey optimization algorithm to update the individual position. The calculation formula is:

[0022] ;

[0023] In the formula, pos ij (t+1) is the position of the i-th individual in the j-th dimension at the t+1-th iteration, j is the individual dimension index, t is the iteration number index, is the position of the i-th individual in the j-th dimension at the t-th iteration, r 1 is a random factor in the range [0,1], lol is the local leader position, r 2 is a random factor in the range [-1,1], is the position of the randomly selected individual in the jth dimension at the tth iteration, rd is a random number in the range [0,1] used to control the individual position update method, and gll is the global leader position;

[0024] Stagnation detection and scout movement unit, when the individual position meets the stagnation condition, the scout movement in the dwarf mongoose optimization algorithm is applied to update the position;

[0025] The stagnation condition specifically refers to the individual position not changing for three consecutive iterations;

[0026] The calculation formula for updating the position by applying the scout movement in the dwarf mongoose optimization algorithm is:

[0027] ;

[0028] In the formula, CF is the control factor used to adjust the degree of exploration. is the average change of individual fitness, is the individual migration value, fit ij(t+1) is the fitness value of the individual position after the update, specifically the fitness value of the position of the i-th individual in the j-th dimension at the t+1-th iteration, fit ij (t) is the fitness value of the current individual position, specifically the fitness value of the position of the i-th individual in the j-th dimension at the t-th iteration;

[0029] The calculation formula of the control factor is:

[0030] ;

[0031] Where, t max is the maximum number of iterations;

[0032] The calculation formula for the average value of individual fitness change is:

[0033] ;

[0034] In the formula, m is the number of individuals, fs i is the change in the fitness value of the i-th individual;

[0035] an iterative calculation unit, repeatedly executing the leader position update unit, the individual position update unit and the stagnation detection and scout movement unit until a maximum number of iterations is reached, and obtaining a global leader position as an optimal feature subset;

[0036] A dwarf mongoose optimization algorithm construction unit combined with spider monkey optimization is constructed by an algorithm initialization unit, a leader position update unit, an individual position update unit, a stagnation detection and scout movement unit and an iterative calculation unit.

[0037] The optimal feature calculation unit for the children's ward selects features of the high-quality data of the children's ward by combining the dwarf mongoose optimization algorithm optimized by spider monkeys, and obtains the optimal features of the children's ward.

[0038] Furthermore, in the vital sign prediction module, a time convolution long-term and short-term parallel model is used to predict vital signs and obtain vital sign prediction information;

[0039] The temporal convolution long-term and short-term parallel model includes an input layer, a temporal convolution long-term and short-term parallel layer, a feature concatenation layer and an output layer;

[0040] The temporal convolution long-short term parallel layer includes a parallel temporal convolution subnet and a long-short term memory subnet;

[0041] The temporal convolutional subnet is used to capture local features and long-term dependencies in the data;

[0042] The long short-term memory subnet is used to capture global features in the data and transfer context information;

[0043] The feature concatenation layer is used to fuse the feature information output by the temporal convolution subnet and the long short-term memory subnet;

[0044] The vital sign prediction module is provided with an input layer construction unit, a time convolution long-term parallel layer construction unit, a feature splicing layer construction unit, an output layer construction unit, a time convolution long-term parallel model construction unit, a model training unit and a prediction result generation unit, including the following contents:

[0045] An input layer construction unit receives the optimal features of the children's ward as input data of the model by constructing the input layer;

[0046] The temporal convolution long-term parallel layer construction unit constructs the temporal convolution long-term parallel layer by setting a temporal convolution subnet and a long short-term memory subnet of a parallel structure in the temporal convolution long-term parallel layer, including the following contents:

[0047] The temporal convolutional subnet is constructed by stacking N convolutional layers in the temporal convolutional subnet, setting causal convolution and dilated convolution in each convolutional layer, setting residual connections between each convolutional layer, and processing the input data through the temporal convolutional subnet to obtain temporal convolution features;

[0048] The causal convolution is used to ensure the order of the time series, and the calculation formula is:

[0049] ;

[0050] In the formula, y1 z is the output of the causal convolution at the zth time step, z is the time step index, k is the convolution kernel weight index, K is the convolution kernel size, and g k is the kth weight of the convolution kernel, x z-k is the value of the input data at time step zk;

[0051] The dilated convolution is used to increase the capture range of long-term dependencies, and the calculation formula is:

[0052] ;

[0053] In the formula, y2 z is the output of the dilated convolution at time step z, d is the dilation rate, The input data is The value of the time step;

[0054] The long short-term memory subnet is constructed, specifically, M long short-term memory layers are stacked in the long short-term memory subnet, wherein the long short-term memory layer includes an input gate, a forget gate, a memory unit and an output gate, and the input data is processed by the long short-term memory subnet to obtain long-term and short-term features;

[0055] The feature concatenation layer construction unit concatenates the temporal convolution features and the long-term and short-term features in the feature dimension through the feature concatenation layer to achieve the fusion of the feature dimension and obtain the temporal convolution long-term and short-term features;

[0056] The output layer constructs a unit, in which the RELU activation function is used to calculate the long-term and short-term features of the temporal convolution to generate the prediction results;

[0057] A temporal convolution long-term and short-term parallel model building unit, which builds a temporal convolution long-term and short-term parallel model through an input layer building unit, a temporal convolution long-term and short-term parallel layer building unit, a feature concatenation layer building unit, an output layer building unit, a temporal convolution long-term and short-term parallel model building unit, a model training unit and a prediction result generating unit;

[0058] The model training unit obtains a vital sign prediction model by training the time convolution long-term and short-term parallel model;

[0059] The prediction result generating unit predicts vital signs through the vital sign prediction model to obtain vital sign prediction information.

[0060] Furthermore, in the prediction result feedback module, a vital sign prediction report is automatically generated based on the vital sign prediction information, and an alarm is issued when there is an abnormality in the vital sign prediction information.

[0061] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0062] (1) In view of the technical problems in the existing vital sign prediction process, the data involved in vital sign prediction is of high dimension and feature redundancy is prone to occur, which leads to model overfitting and inability to effectively predict unknown data, thus affecting the overall practicality and effectiveness of the system. This solution creatively adopts the dwarf mongoose optimization algorithm combined with spider monkey optimization for feature selection. By combining the advantages of the two optimization algorithms, the calculation speed is improved and redundant features are removed, the risk of model overfitting is reduced, and the performance of the prediction model is improved, thereby enhancing the overall practicality and effectiveness of the vital sign prediction system.

[0063] (2) In view of the technical problem that in the existing vital signs prediction process, the vital signs related data have the characteristics of high variability and periodic trends, and a single model is difficult to handle these two characteristics at the same time, resulting in poor model prediction effect and affecting the overall accuracy of the system, this solution creatively adopts the time convolution long-term and short-term parallel model for vital signs prediction, and sets the time convolution subnet and long-term and short-term memory subnet in parallel, so that the model can efficiently capture local details and global trends at the same time, which is more suitable for capturing short-term fluctuations and long-term laws in vital signs related data, thereby improving the model prediction effect and enhancing the overall accuracy of the vital signs prediction system. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A schematic diagram of the structure of a vital sign prediction system for pediatric clinical assistance provided by the present invention;

[0065] Figure 2 It is a flowchart of the feature selection module;

[0066] Figure 3 It is a flowchart of the leader position update unit;

[0067] Figure 4 Schematic diagram of the flow of the vital signs prediction module.

[0068] 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

[0069] 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.

[0070] 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.

[0071] Example 1, see Figure 1, a vital sign prediction system for pediatric clinical assistance provided by the present invention includes: a data preparation module, a data preprocessing module, a feature selection module, a vital sign prediction module and a prediction result feedback module;

[0072] The data preparation module obtains comprehensive data of the children's ward through data preparation, and sends the comprehensive data of the children's ward to the data preprocessing module;

[0073] The data preprocessing module obtains high-quality data of the children's ward through data preprocessing, and sends the high-quality data of the children's ward to the feature selection module;

[0074] The feature selection module uses the dwarf mongoose optimization algorithm combined with spider monkey optimization to perform feature selection to obtain the optimal features of the children's ward, and sends the optimal features of the children's ward to the vital sign prediction module;

[0075] The vital sign prediction module uses a time convolution long-term and short-term parallel model to predict vital signs, obtain vital sign prediction information, and send the vital sign prediction information to the prediction result feedback module;

[0076] The prediction result feedback module automatically generates a vital sign prediction report through prediction result feedback.

[0077] Example 2, see Figure 1 , based on the above embodiment, this embodiment is used in the data preparation module to collect the original data required for vital sign prediction, specifically, to obtain comprehensive data of children's wards by collecting children's vital sign data and ward environment data;

[0078] The child's vital signs data include heart rate, respiratory rate, blood pressure, body temperature, blood oxygen saturation;

[0079] The ward environment data includes ward temperature and humidity.

[0080] Example 3, see Figure 1 ,This embodiment is based on the above embodiment. In the data preprocessing module, the comprehensive data of the children's ward is preprocessed by deleting duplicate values, interpolating missing values ​​and standardizing data to obtain high-quality data of the children's ward.

[0081] Example 4, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In the feature selection module, an algorithm initialization unit, a leader position update unit, an individual position update unit, a stagnation detection and scout movement unit, an iterative calculation unit, a dwarf mongoose optimization algorithm construction unit combined with spider monkey optimization, and a children's ward optimal feature calculation unit are provided, including the following contents:

[0082] An algorithm initialization unit randomly initializes individual positions in the feature space of high-quality data in the children's ward, sets the number of individuals and the maximum number of iterations, and uses the prediction accuracy of the vital sign prediction model as a fitness function. The individual positions are used to represent feature subsets;

[0083] The leader position update unit updates the local leader position according to the relative fitness value and updates the global leader position according to the fitness value, including the following:

[0084] The fitness function is used to evaluate the fitness value of each individual's current position and calculate the relative fitness value. The calculation formula of the relative fitness value is:

[0085] ;

[0086] In the formula, is the relative fitness value, i is the individual index, used to represent the current individual, fit i is the fitness value of the i-th individual, n is the number of individuals in the current individual's domain;

[0087] Update the local leader position, specifically, in each individual area, select the individual position with the highest relative fitness value as the local leader position;

[0088] Update the global leader position, specifically, select the individual position with the highest fitness value among all individuals as the global leader position;

[0089] The individual position update unit uses the spider monkey optimization algorithm to update the individual position. The calculation formula is:

[0090] ;

[0091] In the formula, pos ij (t+1) is the position of the i-th individual in the j-th dimension at the t+1-th iteration, j is the individual dimension index, t is the iteration number index, is the position of the i-th individual in the j-th dimension at the t-th iteration, r 1 is a random factor in the range [0,1], lol is the local leader position, r 2 is a random factor in the range [-1,1], is the position of the randomly selected individual in the jth dimension at the tth iteration, rd is a random number in the range [0,1] used to control the individual position update method, and gll is the global leader position;

[0092] Stagnation detection and scout movement unit, when the individual position meets the stagnation condition, the scout movement in the dwarf mongoose optimization algorithm is applied to update the position;

[0093] The stagnation condition specifically refers to the individual position not changing for three consecutive iterations;

[0094] The calculation formula for updating the position by applying the scout movement in the dwarf mongoose optimization algorithm is:

[0095] ;

[0096] In the formula, CF is the control factor used to adjust the degree of exploration. is the average change of individual fitness, is the individual migration value, fit ij (t+1) is the fitness value of the individual position after the update, specifically the fitness value of the position of the i-th individual in the j-th dimension at the t+1-th iteration, fit ij (t) is the fitness value of the current individual position, specifically the fitness value of the position of the i-th individual in the j-th dimension at the t-th iteration;

[0097] The calculation formula of the control factor is:

[0098] ;

[0099] Where, t max is the maximum number of iterations;

[0100] The calculation formula for the average value of individual fitness change is:

[0101] ;

[0102] In the formula, m is the number of individuals, fs i is the change in the fitness value of the i-th individual;

[0103] an iterative calculation unit, repeatedly executing the leader position update unit, the individual position update unit and the stagnation detection and scout movement unit until a maximum number of iterations is reached, and obtaining a global leader position as an optimal feature subset;

[0104] A dwarf mongoose optimization algorithm construction unit combined with spider monkey optimization is constructed by an algorithm initialization unit, a leader position update unit, an individual position update unit, a stagnation detection and scout movement unit and an iterative calculation unit.

[0105] The optimal feature calculation unit for children's wards selects the features of high-quality data of children's wards by combining the dwarf mongoose optimization algorithm optimized by spider monkeys, and obtains the optimal features of children's wards;

[0106] Table 1 is a performance comparison of the vital sign prediction model before and after feature selection in Example 4 of the present invention. As shown in the table, the F1-score is the harmonic mean of the precision rate and the recall rate. The higher the F1-score and the prediction accuracy, the more representative the selected features are.

[0107] Table 1 Performance comparison of vital sign prediction models before and after feature selection

[0108]

[0109] By performing the above operations, in order to solve the technical problem that in the existing vital signs prediction process, the data dimension involved in the vital signs prediction is high and feature redundancy is prone to occur, which leads to model overfitting and inability to effectively predict unknown data, thereby affecting the overall practicality and effectiveness of the system, this solution creatively adopts the dwarf mongoose optimization algorithm combined with spider monkey optimization for feature selection. By combining the advantages of the two optimization algorithms, the calculation speed is improved and redundant features are removed, the risk of model overfitting is reduced, and the performance of the prediction model is improved, thereby enhancing the overall practicality and effectiveness of the vital signs prediction system.

[0110] Example 5, see Figure 1 and Figure 4 , based on the above embodiment, in the vital sign prediction module, this embodiment adopts the time convolution long-term and short-term parallel model to perform vital sign prediction and obtain vital sign prediction information;

[0111] The temporal convolution long-term and short-term parallel model includes an input layer, a temporal convolution long-term and short-term parallel layer, a feature concatenation layer and an output layer;

[0112] The temporal convolution long-short term parallel layer includes a parallel temporal convolution subnet and a long-short term memory subnet;

[0113] The temporal convolutional subnet is used to capture local features and long-term dependencies in the data;

[0114] The long short-term memory subnet is used to capture global features in the data and transfer context information;

[0115] The feature concatenation layer is used to fuse the feature information output by the temporal convolution subnet and the long short-term memory subnet;

[0116] The vital sign prediction module is provided with an input layer construction unit, a time convolution long-term parallel layer construction unit, a feature splicing layer construction unit, an output layer construction unit, a time convolution long-term parallel model construction unit, a model training unit and a prediction result generation unit, including the following contents:

[0117] An input layer construction unit receives the optimal features of the children's ward as input data of the model by constructing the input layer;

[0118] The temporal convolution long-term parallel layer construction unit constructs the temporal convolution long-term parallel layer by setting a temporal convolution subnet and a long short-term memory subnet of a parallel structure in the temporal convolution long-term parallel layer, including the following contents:

[0119] The temporal convolutional subnet is constructed by stacking N convolutional layers in the temporal convolutional subnet, setting causal convolution and dilated convolution in each convolutional layer, setting residual connections between each convolutional layer, and processing the input data through the temporal convolutional subnet to obtain temporal convolution features;

[0120] The causal convolution is used to ensure the order of the time series, and the calculation formula is:

[0121] ;

[0122] In the formula, y1 z is the output of the causal convolution at the zth time step, z is the time step index, k is the convolution kernel weight index, K is the convolution kernel size, and g k is the kth weight of the convolution kernel, x z-k is the value of the input data at time step zk;

[0123] The dilated convolution is used to increase the capture range of long-term dependencies, and the calculation formula is:

[0124] ;

[0125] In the formula, y2 z is the output of the dilated convolution at time step z, d is the dilation rate, The input data is The value of the time step;

[0126] The long short-term memory subnet is constructed, specifically, M long short-term memory layers are stacked in the long short-term memory subnet, wherein the long short-term memory layer includes an input gate, a forget gate, a memory unit and an output gate, and the input data is processed by the long short-term memory subnet to obtain long-term and short-term features;

[0127] The feature concatenation layer construction unit concatenates the temporal convolution features and the long-term and short-term features in the feature dimension through the feature concatenation layer to achieve the fusion of the feature dimension and obtain the temporal convolution long-term and short-term features;

[0128] The output layer constructs a unit, in which the RELU activation function is used to calculate the long-term and short-term features of the temporal convolution to generate the prediction results;

[0129] A temporal convolution long-term and short-term parallel model building unit, which builds a temporal convolution long-term and short-term parallel model through an input layer building unit, a temporal convolution long-term and short-term parallel layer building unit, a feature concatenation layer building unit, an output layer building unit, a temporal convolution long-term and short-term parallel model building unit, a model training unit and a prediction result generating unit;

[0130] The model training unit obtains a vital sign prediction model by training the time convolution long-term and short-term parallel model;

[0131] A prediction result generating unit performs a vital sign prediction through a vital sign prediction model to obtain vital sign prediction information;

[0132] By performing the above operations, in order to address the technical problem that in the existing vital signs prediction process, the vital signs-related data have the characteristics of high variability and periodic trends, and a single model is difficult to handle these two characteristics at the same time, resulting in poor model prediction effect and affecting the overall accuracy of the system, this solution creatively adopts the time convolution long-term and short-term parallel model for vital signs prediction, and sets the time convolution subnet and long-term and short-term memory subnet in parallel, so that the model can efficiently capture local details and global trends at the same time, and is more suitable for capturing short-term fluctuations and long-term rules in vital signs-related data, thereby improving the model prediction effect and enhancing the overall accuracy of the vital signs prediction system.

[0133] Example 6, see Figure 1 This embodiment is based on the above embodiment. In the prediction result feedback module, a vital sign prediction report is automatically generated according to the vital sign prediction information. When there is an abnormality in the vital sign prediction information, an alarm is issued.

[0134] 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.

[0135] 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.

[0136] 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 vital sign prediction system for pediatric clinical assistance, characterized by: It includes data preparation module, data preprocessing module, feature selection module, vital sign prediction module and prediction result feedback module; The data preparation module is used to collect the original data required for vital sign prediction, specifically to obtain comprehensive data of the children's ward by collecting the children's vital sign data and ward environment data, and to send the comprehensive data of the children's ward to the data preprocessing module; The child's vital signs data include heart rate, respiratory rate, blood pressure, body temperature and blood oxygen saturation; the ward environment data include ward temperature and humidity; The data preprocessing module obtains high-quality data of the children's ward through data preprocessing, and sends the high-quality data of the children's ward to the feature selection module; The feature selection module uses the dwarf mongoose optimization algorithm combined with spider monkey optimization to perform feature selection to obtain the optimal features of the children's ward, and sends the optimal features of the children's ward to the vital sign prediction module; In the feature selection module, there are an algorithm initialization unit, a leader position update unit, an individual position update unit, a stagnation detection and scout movement unit, an iterative calculation unit, a dwarf mongoose optimization algorithm construction unit combined with spider monkey optimization, and a children's ward optimal feature calculation unit, including the following: An algorithm initialization unit randomly initializes individual positions in the feature space of high-quality data in the children's ward, sets the number of individuals and the maximum number of iterations, and uses the prediction accuracy of the vital sign prediction model as a fitness function. The individual positions are used to represent feature subsets; The leader position update unit updates the local leader position according to the relative fitness value and updates the global leader position according to the fitness value, including the following: The fitness function is used to evaluate the fitness value of each individual's current position and calculate the relative fitness value. The calculation formula of the relative fitness value is: ； In the formula, is the relative fitness value, i is the individual index, used to represent the current individual, fit i is the fitness value of the i-th individual, n is the number of individuals in the current individual's domain; Update the local leader position, specifically, in each individual area, select the individual position with the highest relative fitness value as the local leader position; Update the global leader position, specifically, select the individual position with the highest fitness value among all individuals as the global leader position; The individual position update unit uses the spider monkey optimization algorithm to update the individual position. The calculation formula is: ； In the formula, pos ij (t+1) is the position of the i-th individual in the j-th dimension at the t+1-th iteration, j is the individual dimension index, t is the iteration number index, is the position of the ith individual in the jth dimension at the tth iteration, r1 is a random factor in the range [0,1], lol is the local leader position, r2 is a random factor in the range [-1,1], is the position of the randomly selected individual in the jth dimension at the tth iteration, rd is a random number in the range [0,1] used to control the individual position update method, and gll is the global leader position; Stagnation detection and scout movement unit, when the individual position meets the stagnation condition, the scout movement in the dwarf mongoose optimization algorithm is applied to update the position; The stagnation condition specifically refers to the individual position not changing for three consecutive iterations; The calculation formula for updating the position by applying the scout movement in the dwarf mongoose optimization algorithm is: ； In the formula, CF is the control factor used to adjust the degree of exploration. is the average change of individual fitness, is the individual migration value; The calculation formula of the control factor is: ； Where, t max is the maximum number of iterations; The calculation formula for the average value of individual fitness change is: ； In the formula, m is the number of individuals, fs i is the change in the fitness value of the i-th individual; an iterative calculation unit, repeatedly executing the leader position update unit, the individual position update unit and the stagnation detection and scout movement unit until a maximum number of iterations is reached, and obtaining a global leader position as an optimal feature subset; A dwarf mongoose optimization algorithm construction unit combined with spider monkey optimization is constructed by an algorithm initialization unit, a leader position update unit, an individual position update unit, a stagnation detection and scout movement unit and an iterative calculation unit. The optimal feature calculation unit for children's wards selects the features of high-quality data of children's wards by combining the dwarf mongoose optimization algorithm optimized by spider monkeys, and obtains the optimal features of children's wards; The vital sign prediction module uses a time convolution long-term and short-term parallel model to predict vital signs, obtain vital sign prediction information, and send the vital sign prediction information to the prediction result feedback module; The temporal convolution long-term and short-term parallel model includes an input layer, a temporal convolution long-term and short-term parallel layer, a feature concatenation layer and an output layer; The temporal convolution long-short term parallel layer includes a parallel temporal convolution subnet and a long-short term memory subnet; The temporal convolutional subnet is used to capture local features and long-term dependencies in the data; The long short-term memory subnet is used to capture global features in the data and transfer context information; The feature concatenation layer is used to fuse the feature information output by the temporal convolution subnet and the long short-term memory subnet; The prediction result feedback module automatically generates a vital sign prediction report through prediction result feedback; In the vital sign prediction module, there are input layer construction units, time convolution long-term parallel layer construction units, feature splicing layer construction units, output layer construction units, time convolution long-term parallel model construction units, model training units and prediction result generation units, including the following contents: An input layer construction unit receives the optimal features of the children's ward as input data of the model by constructing the input layer; A temporal convolution long-term and short-term parallel layer construction unit is used to construct a temporal convolution long-term and short-term parallel layer by setting a temporal convolution subnet and a long short-term memory subnet of a parallel structure in the temporal convolution long-term and short-term parallel layer; The feature concatenation layer construction unit concatenates the temporal convolution features and the long-term and short-term features in the feature dimension through the feature concatenation layer to achieve the fusion of the feature dimension and obtain the temporal convolution long-term and short-term features; The output layer constructs a unit, in which the RELU activation function is used to calculate the long-term and short-term features of the temporal convolution to generate the prediction results; A temporal convolution long-term and short-term parallel model building unit, which builds a temporal convolution long-term and short-term parallel model through an input layer building unit, a temporal convolution long-term and short-term parallel layer building unit, a feature concatenation layer building unit, an output layer building unit, a temporal convolution long-term and short-term parallel model building unit, a model training unit and a prediction result generating unit; The temporal convolution long-term and short-term parallel layer construction unit includes the following contents: The temporal convolutional subnet is constructed by stacking N convolutional layers in the temporal convolutional subnet, setting causal convolution and dilated convolution in each convolutional layer, setting residual connections between each convolutional layer, and processing the input data through the temporal convolutional subnet to obtain temporal convolution features; The long short-term memory subnet is constructed, specifically, M long short-term memory layers are stacked in the long short-term memory subnet, wherein the long short-term memory layer includes an input gate, a forget gate, a memory unit, and an output gate, and the input data is processed by the long short-term memory subnet to obtain long-term and short-term features; The model training unit obtains a vital sign prediction model by training the time convolution long-term and short-term parallel model; The prediction result generating unit predicts vital signs through the vital sign prediction model to obtain vital sign prediction information.

2. A vital sign prediction system for pediatric clinical assistance according to claim 1, characterized in that: In the time convolution long-term and short-term parallel layer building unit, the causal convolution is used to ensure the order of the time series, and the calculation formula is: ； In the formula, y1 z is the output of the causal convolution at the zth time step, z is the time step index, k is the convolution kernel weight index, K is the convolution kernel size, and g k is the kth weight of the convolution kernel, x z-k is the value of the input data at time step zk; The dilated convolution is used to increase the capture range of long-term dependencies, and the calculation formula is: ； In the formula, y2 z is the output of the dilated convolution at time step z, d is the dilation rate, The input data is The value of the time step.

3. A vital sign prediction system for pediatric clinical assistance according to claim 2, characterized in that: In the prediction result feedback module, a vital sign prediction report is automatically generated based on the vital sign prediction information, and an alarm is issued when there is an abnormality in the vital sign prediction information.

4. A vital sign prediction system for pediatric clinical assistance according to claim 3, characterized in that: In the data preprocessing module, the comprehensive data of the children's ward are preprocessed through duplicate value deletion, missing value interpolation and data standardization to obtain high-quality data of the children's ward.

Images