A perioperative stress recognition method and system based on HRV intelligent monitoring

Abstract

The application provides a perioperative stress identification method and system based on HRV intelligent monitoring, and relates to the technical field of medical data processing.The method comprises the following steps: determining a segmented period in the perioperative period in which a monitoring period is located; determining a plurality of HRV data and physiological index data, and first input vectors and second input vectors thereof; determining period encoding information according to the segmented period and the starting time of the monitoring period; and further determining a stress index through a stress identification model.According to the application, a plurality of signals of a patient in a perioperative period can be acquired through a wearable physiological signal acquisition device, and the characteristics that an HRV signal can reflect the sympathetic nerve condition are utilized to identify whether the patient has a stress reaction in each segmented period, and a plurality of physiological index data are utilized for auxiliary judgment to improve the accuracy and confidence of identification, and the plurality of acquired data can be continuously acquired and updated to improve the timeliness of identification.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical data processing, and particularly relates to a perioperative stress identification method and system based on HRV intelligent monitoring. BACKGROUND

[0002] The perioperative period is a high-risk stage in clinical practice, and patients are prone to significant stress reactions, which poses challenges to preoperative preparation, intraoperative anesthesia management, and postoperative recovery. Current clinical routine monitoring methods usually include monitoring heart rate, blood pressure, heart rate variability (HRV), etc. These monitoring methods have problems such as response delay and low monitoring accuracy, making it difficult to reflect the state of the autonomic nervous system in a timely manner. SUMMARY

[0003] The present application provides a perioperative stress identification method and system based on HRV intelligent monitoring, which can solve the technical problems of response delay and low monitoring accuracy in monitoring stress reactions in related technologies.

[0004] According to a first aspect of the present application, a perioperative stress identification method based on HRV intelligent monitoring is provided, comprising:

[0005] At the beginning of the perioperative period, a physiological signal acquisition device is worn by the patient;

[0006] Determine the segmented period in the perioperative period in which the current monitoring period is located;

[0007] Determine the multiple HRV data of the current monitoring period and the physiological index data at multiple time points within the current monitoring period through the physiological signal acquisition device;

[0008] Obtain a first input vector of a stress identification model according to the multiple HRV data;

[0009] Obtain a second input vector corresponding to each physiological index data according to the physiological index data at multiple time points within the current monitoring period;

[0010] Determine time period encoding information according to the segmented period and the start time of the monitoring period;

[0011] Determine the stress index of the patient according to the time period encoding information, the first input vector, the second input vector, and the stress identification model;

[0012] Determine stress prompt information according to the stress index of the patient.

[0013] According to the present application, a first input vector of a stress identification model is obtained according to multiple HRV data, comprising:

[0014] Assemble the plurality of HRV data into an HRV data vector;

[0015] Process the HRV data vector through a first multi-layer perception network level of the stress recognition model to obtain a first input vector;

[0016] According to the physiological index data at a plurality of time points in the monitoring period, obtain a second input vector corresponding to each kind of physiological index data, comprising:

[0017] Assemble the physiological index data at a plurality of time points in the monitoring period into a physiological index data sequence;

[0018] Process the physiological index data sequence through a 1D convolution network level of the stress recognition model to obtain a second input vector corresponding to the physiological index data;

[0019] According to the segment period and the starting time point of the monitoring period, determine segment encoding information, comprising:

[0020] Determine the first serial number of the starting time point of the monitoring period in the segment period;

[0021] Combine the category information of the segment period and the first serial number to obtain a segment vector;

[0022] Process the segment vector through a second multi-layer perception network level of the stress recognition model to obtain the segment encoding information.

[0023] According to the present application, according to the segment encoding information, the first input vector, the second input vector and the stress recognition model, determine the stress index of the patient, comprising:

[0024] The first input vector and the second input vector of the plurality of physiological index data are respectively taken as the input vector of the node of the graph structure;

[0025] Process the segment encoding information through a third multi-layer perception network level of the stress recognition model to obtain an attention mechanism corresponding to the monitoring period;

[0026] According to the input vector of each node and the attention mechanism corresponding to the monitoring period, obtain the connection weight between each node;

[0027] Process the input vector of the node of the graph structure and the connection weight between each node through a graph neural network sub-model of the stress recognition model to obtain the output vector of each node;

[0028] According to the segment encoding information and the output vector of each node, obtain the first output vector of the current monitoring period;

[0029] According to the time period encoding information and the first output vector, a stress index of the patient is determined.

[0030] According to the time period encoding information and the output vector of each node, the first output vector of the current monitoring period is obtained, including:

[0031] The output vectors of the nodes are combined to obtain a first output matrix;

[0032] The time period encoding information is processed through a fourth multi-layer perception network level of the stress recognition model to obtain a time period mapping vector;

[0033] The first output matrix is multiplied by the time period mapping vector to obtain the first output vector.

[0034] According to the time period encoding information and the first output vector, a stress index of the patient is determined, including:

[0035] According to the time period encoding information, it is determined whether the monitoring period is the first monitoring period of the current segmented period;

[0036] If the monitoring period is the first monitoring period of the current segmented period, the first output vector is spliced with a zero vector, and the spliced vector is taken as a stress state vector of the current monitoring period, and the stress state vector is processed through a fifth multi-layer perception network level of the stress recognition model to obtain the stress index of the patient;

[0037] If the monitoring period is not the first monitoring period of the current segmented period, the time period encoding information is processed through a sixth multi-layer perception network level of the stress recognition model to obtain a state transition matrix;

[0038] The first output vector of the previous monitoring period is processed through the state transition matrix to obtain a state transition vector;

[0039] The state transition vector is spliced with the first output vector of the current monitoring period to obtain a stress state vector of the current monitoring period;

[0040] The stress state vector of the current monitoring period is processed through the fifth multi-layer perception network level of the stress recognition model to obtain the stress index of the patient.

[0041] According to the present application, the training step of the stress recognition model includes:

[0042] First sample input vectors, second sample input vectors and sample time period encoding information of sample patients who have stress and sample patients who have no stress in each segmented period of the perioperative period are obtained respectively;

[0043] Adjust at least one of preset type of HRV data and physiological index data of the sample patient in each segmented period to obtain a first sample input vector, a second sample input vector and sample period encoding information of the pseudo sample patient;

[0044] Select the sample period encoding information of the pseudo sample patient, input a third multi-layer perception network level of the stress recognition model, and obtain a training attention mechanism;

[0045] Process the training attention mechanism through a first full connection layer and a first activation layer to obtain an adjustment type recognition result;

[0046] Determine an attention loss function according to the adjustment type recognition result and the HRV data and the physiological index data that are adjusted;

[0047] Select the first sample input vector, the second sample input vector and the sample period encoding information of the arbitrary sample patient or the pseudo sample patient, and obtain a corresponding training period mapping vector and a training state transition matrix;

[0048] Process the training state transition matrix through a second full connection layer and a second activation layer to obtain a training period vector, and splice the training period vector with the training period mapping vector to obtain a training period recognition vector;

[0049] Process the training period recognition vector through a third full connection layer and a third activation layer to obtain a training period recognition result;

[0050] Determine a segmentation recognition loss function according to the training period recognition result, a segmented period corresponding to the sample period encoding information and a first sequence number of a start time of a monitoring cycle corresponding to the sample period encoding information in the segmented period;

[0051] Process the first sample input vector, the second sample input vector and the sample period encoding information of the selected sample patient or the pseudo sample patient through the stress recognition model to obtain a training stress indicator;

[0052] According to the segmented period corresponding to the sample period encoding information and the first sequence number of the start time of the monitoring cycle corresponding to the sample period encoding information in the segmented period, the training stress indicator and the labeled information of the sample patient or the pseudo sample patient, obtain a stress recognition loss function;

[0053] Determine a loss function of the stress recognition model according to the stress recognition loss function, the segmentation recognition loss function and the attention loss function;

[0054] Train the stress recognition model according to the loss function of the stress recognition model to obtain a trained stress recognition model.

[0055] According to the application, the attention loss function is determined according to the adjustment type identification result and the HRV data and physiological index data subjected to adjustment, and comprises:

[0056] According to the formula ;

[0057] The attention loss function is determined , wherein, is the actual probability of the i th HRV data subjected to adjustment, is the probability of the i th HRV data subjected to adjustment determined according to the adjustment type identification result, is the actual probability of the j th physiological index data subjected to adjustment, is the probability of the j th physiological index data subjected to adjustment determined according to the adjustment type identification result, is the number of types of the HRV data, is the number of types of the physiological index data, i≤ , j≤ , and i, , j, are all positive integers.

[0058] According to the application, the segmentation identification loss function is determined according to the training period identification result, the first sequence number of the start moment of the monitoring period corresponding to the sample period coding information in the segmentation period corresponding to the sample period coding information, and comprises:

[0059] According to the formula ;

[0060] The segmentation identification loss function is determined , wherein, is the representation information of the segmentation period determined according to the training period identification result, is the representation information of the segmentation period corresponding to the sample period coding information, is the first sequence number determined according to the training period identification result, is the first sequence number corresponding to the sample period coding information, and are preset weights, and .

[0061] According to the application, the stress identification loss function is obtained by training the stress index and the label information of the sample patient or the pseudo sample patient according to the first sequence number of the start moment of the monitoring period corresponding to the sample period coding information in the segmentation period corresponding to the sample period coding information, and comprises:

[0062] According to the formula ;

[0063] Stress recognition loss function wherein, is a probability of the monitoring period appearing a stress condition according to the labeled information, is a training stress indicator of the monitoring period, is the first serial number, is a total number of periods in the segmented period to which the monitoring period belongs.

[0064] According to a second aspect of the present application, a perioperative stress recognition system based on HRV intelligent monitoring is provided, comprising:

[0065] A wearing module is configured to wear a physiological signal acquisition device for a patient at the beginning of a perioperative period;

[0066] A segmented period module is configured to determine a segmented period in the perioperative period in which a current monitoring period is located;

[0067] An acquisition module is configured to determine a plurality of HRV data of the current monitoring period and physiological indicator data at a plurality of time points in the current monitoring period through the physiological signal acquisition device;

[0068] A first input vector module is configured to obtain a first input vector of a stress recognition model according to the plurality of HRV data;

[0069] A second input vector module is configured to obtain a second input vector corresponding to each physiological indicator data according to the physiological indicator data at the plurality of time points in the current monitoring period;

[0070] A time period encoding information module is configured to determine time period encoding information according to the segmented period and a starting time point of the monitoring period;

[0071] A stress indicator module is configured to determine a stress indicator of the patient according to the time period encoding information, the first input vector, the second input vector and the stress recognition model;

[0072] A stress prompt information module is configured to determine stress prompt information according to the stress indicator of the patient.

[0073] By adopting the above technical solutions, the present application can achieve the following technical effects:

[0074] According to the present application, the physiological signal acquisition device can be worn to obtain various signals of the patient during the perioperative period, and the HRV signal can reflect the characteristics of the sympathetic nervous condition, identify whether the patient has a stress reaction in each segmented period, and use various physiological indicator data for auxiliary judgment to improve the accuracy and confidence of the identification, and the various data collected can be continuously obtained and updated to improve the timeliness of the identification. During the determination of the stress indicator of the patient, the important HRV data type and the important physiological indicator data type can be highlighted through the attention mechanism, the weight of which is improved, and based on the differences in the judgment criteria of each monitoring period, the attention mechanism corresponding to the monitoring period can be determined based on the time period coding information, so as to enhance the features of the important data type, thereby improving the accuracy of determining the stress indicator. When determining whether the patient has stress, the segmented period in which the monitoring period is located and the serial number of the monitoring period in the segmented period are combined, so that the judgment mode and the judgment standard of different monitoring periods are considered during the judgment, the accuracy and adaptability of the judgment are improved, and the features of the previous monitoring period can be combined for judgment, and the features of the previous monitoring period can be aligned during the judgment, and the features are compared after alignment, which improves the objectivity and accuracy of the comparison, thereby improving the accuracy of the stress indicator. When determining the attention loss function, the first fully connected layer and the first activation layer are set to determine the HRV data and the physiological indicator data type enhanced by the training attention mechanism, and the training attention mechanism can enhance the features of the important type of HRV data and physiological indicator data during the training, which improves the enhancement ability of the training attention mechanism, and further improves the accuracy of the stress recognition model in determining the stress indicator. When determining the segmented recognition loss function, the second fully connected layer and the second activation layer and the third fully connected layer and the third activation layer are set to predict the segmented period and the first serial number corresponding to the training period mapping vector and the training state transition matrix, so as to set the segmented recognition loss function to improve the accuracy of the segmented period and the first serial number predicted by the model during the training process, and further improve the accuracy of the attention mechanism, the period mapping vector and the state transition matrix. When determining the stress recognition loss function, the weight of the cross-entropy loss function corresponding to the training stress indicator of each monitoring period can be set during the training process, and the weight can be set based on the influence of stress on the operation and the corresponding segmented period, which can improve the pertinence and training efficiency of the training. BRIEF DESCRIPTION OF DRAWINGS

[0075] Figure 1 An exemplary flowchart of a perioperative stress recognition method based on HRV intelligent monitoring according to an embodiment of the present application is shown;

[0076] Figure 2 An exemplary diagram for obtaining a stress indicator according to an embodiment of the present application is shown;

[0077] Figure 3A block diagram of a perioperative stress identification system based on HRV intelligent monitoring according to an embodiment of the present application is exemplarily shown. DETAILED DESCRIPTION

[0078] The technical solutions of the present application are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0079] Figure 1 A flowchart of a perioperative stress identification method based on HRV intelligent monitoring according to an embodiment of the present application is exemplarily shown, and the method comprises:

[0080] Step S1, at the beginning of the perioperative period, wearing a physiological signal acquisition device for the patient;

[0081] Step S2, determining the segmented period in the perioperative period in which the current monitoring period is located;

[0082] Step S3, determining, by the physiological signal acquisition device, a plurality of HRV data of the current monitoring period, and physiological index data at a plurality of time points in the current monitoring period;

[0083] Step S4, obtaining a first input vector of a stress identification model according to the plurality of HRV data;

[0084] Step S5, obtaining a second input vector corresponding to each physiological index data according to the physiological index data at a plurality of time points in the current monitoring period;

[0085] Step S6, determining period encoding information according to the segmented period and the starting time point of the monitoring period;

[0086] Step S7, determining a stress index of the patient according to the period encoding information, the first input vector, the second input vector, and the stress identification model;

[0087] Step S8, determining stress prompt information according to the stress index of the patient.

[0088] The perioperative stress identification method based on HRV intelligent monitoring according to the embodiment of the present application can obtain a plurality of signals of the patient in the perioperative period through the wearable physiological signal acquisition device, utilize the characteristic that the HRV signal can reflect the sympathetic nerve condition, identify whether the patient has a stress reaction in each segmented period, utilize a plurality of physiological index data for auxiliary judgment, improve the accuracy and confidence of the identification, and continuously obtain and update the plurality of collected data, and improve the timeliness of the identification.

[0089] According to one embodiment of the present application, in step S1, the physiological signal acquisition device can be used to acquire a plurality of HRV data, such as RMSSD data (Root Mean Square of Successive Differences), LF data (low frequency power), HF (high frequency power), LF / HF, etc., and a set of HRV data can be obtained in each monitoring period. In addition, the physiological signal acquisition device can also acquire a plurality of physiological index data at a plurality of time points in the monitoring period, such as electrocardiogram data, blood pressure data, blood oxygen data, body temperature data, etc. The length of each monitoring period can be 10 minutes, 20 minutes, etc., and the present application does not limit this.

[0090] According to one embodiment of the present application, in step S2, the perioperative period can be divided into three segmented periods, i.e., a preoperative period, an intraoperative period, and a postoperative period. The segmented period in which the current monitoring period is located can be determined, for example, the type of segmented period can be manually input, or it can be determined automatically based on the operation time. For example, the operation time can be input as a certain time on a certain date. If the current monitoring period is before the time on the date, the current monitoring period belongs to the preoperative period. If the current monitoring period is after the time on the date, and the patient is determined to be in the operating room based on the positioning device, the current monitoring period belongs to the intraoperative period. If the current monitoring period is after the patient leaves the operating room until a certain time (for example, a certain time on the 7th day after the operation), the current monitoring period belongs to the postoperative period. Further, when the segmented period changes, the monitoring period can be restarted, for example, when the above operation time is determined to be reached and the patient enters the operating room, even if the previous monitoring period (belonging to the preoperative period) has not ended, it can be determined that the previous monitoring period ends and the first monitoring period of the intraoperative period begins.

[0091] According to one embodiment of the present application, in step S3, each monitoring period can include a plurality of time points, and a plurality of HRV data can be calculated based on the electrocardiogram data and other data collected at the plurality of time points, thereby obtaining the HRV data of the monitoring period.

[0092] Figure 2 An exemplary diagram for obtaining a stress index according to an embodiment of the present application is shown.

[0093] According to one embodiment of the present application, in step S4, the first input vector can be used to describe a plurality of features of the HRV data. According to the plurality of HRV data, the first input vector of the stress recognition model is obtained, including: grouping the plurality of HRV data into an HRV data vector; processing the HRV data vector through a first multi-layer perception network level of the stress recognition model to obtain the first input vector. In an example, the first multi-layer perception network level can include a plurality of fully connected layers and activation layers (for example, a level processed using a RELU activation function), and can process the HRV data vector to obtain higher-dimensional feature information, i.e., the first input vector, to more comprehensively express the features of the HRV data.

[0094] According to one embodiment of the present application, in step S5, according to the physiological index data at a plurality of time points in the monitoring period, the second input vector corresponding to each physiological index data is obtained, including: grouping the physiological index data at a plurality of time points in the monitoring period into a physiological index data sequence; processing the physiological index data sequence through a 1D convolution network level of the stress recognition model to obtain the second input vector corresponding to the physiological index data. The 1D convolution network level can process the physiological index data sequence, wherein the 1D convolution network level can include a plurality of 1D convolution levels, and each level can include a plurality of convolution kernels. After the above processing, the second input vector can be obtained, which can be used to describe a plurality of features of the physiological index data sequence, such as time domain features, frequency domain features, change features, etc.

[0095] According to one embodiment of the present application, in step S6, according to the segment period and the start time of the monitoring period, the time period encoding information is determined, including: determining the first sequence number of the start time of the monitoring period in the segment period; combining the category information of the segment period and the first sequence number to obtain a time period vector; and processing the time period vector through the second multi-layer perception network level of the stress recognition model to obtain the time period encoding information. The judgment standard of the stress reaction is different when the segment period to which the current monitoring period belongs is different. For example, in the preoperative period, the patient may have anxiety, and the closer to the operation time, the more serious the anxiety, which may cause abnormal ECG signals. In the intraoperative period, the patient may be in a state of anesthesia, and the HRV data and multiple physiological indicators can be monitored at any time to monitor the physical condition. In the postoperative period, the patient may have abnormal HRV data due to postoperative pain, and as the wound heals, the pain will subside, and the abnormal HRV data will be relieved. Therefore, the manifestation form and judgment method of the stress reaction may be different at different times in different segment periods. Therefore, the category information of the segment period (i.e., which segment period the current monitoring period belongs to) and the first sequence number (i.e., a plurality of monitoring periods can be included in each segment period, and the first sequence number can represent the sequence number of the current monitoring period in the plurality of monitoring periods) can be combined to obtain a time period vector, and then the time period vector is processed through the second multi-layer perception network level to obtain the time period encoding information, that is, the feature description of the time period vector is obtained through the multi-layer fully connected layer and the activation layer, that is, the time period encoding information.

[0096] According to one embodiment of the present application, the first input vector represents the features of multiple aspects of the HRV data, and the second input vector of each physiological indicator data represents the features of multiple aspects of the physiological indicator data. The HRV data and the multiple physiological indicators are correlated with each other, and can be used as one of the reference bases for judging the stress reaction. Therefore, the relationship between the HRV data and the multiple physiological indicators can be determined through the graph neural network sub-model, and the features of the HRV data and the features of the multiple physiological indicator data can be fused, so that a comprehensive feature representation can be obtained, and the accuracy of the stress reaction judgment can be improved.

[0097] According to one embodiment of the present application, in the graph structure, the HRV data and each physiological indicator data can be taken as a node, the first input vector and the second input vector can be taken as the input vectors of the respective nodes, and the connection weights between the respective nodes can be determined through the attention mechanism. The higher the connection weight is, the closer the relationship between the two nodes is, and the more important the type of physiological indicator data or the type of HRV data corresponding to the nodes is for judging stress. In an example, the input vectors of the two nodes can be spliced, and the spliced vectors can be processed through the attention mechanism (for example, the attention mechanism is a coefficient matrix) (for example, a vector in the coefficient matrix is multiplied by the spliced vector) to obtain the connection weights between the respective nodes. Moreover, since the judgment criteria and the judgment manner can be different for each monitoring period, the types of indicators that are focused on for monitoring can also be different for each monitoring period, and therefore, the period encoding information can be processed through a third multi-layer perception network level to obtain an attention mechanism corresponding to the monitoring period, so as to enhance the features of the important types of physiological indicator data or HRV data in the monitoring period corresponding to the period encoding information. In subsequent processing, each node can output comprehensive feature information containing its own features and the features of other nodes. The higher the connection weight between the node and other nodes is, or the higher the weight of the node itself is, the more the feature information of other nodes or the feature information of the node itself can be contained in the comprehensive feature information. In the process of judging whether stress occurs by using the comprehensive feature information, the feature information with a higher weight is more important for judging whether stress occurs, and therefore, the parameters of the attention mechanism can be trained in the training process, so as to determine the importance of each feature and further determine the connection weights of the feature information of the respective nodes, so as to highlight the weights of the important types of physiological indicator data or HRV data.

[0098] According to one embodiment of the present application, after the above connection weights are obtained, the input vectors of the respective nodes and the input vectors of other nodes can be weighted and summed through the connection weights to obtain the weighted sum feature vectors of the respective nodes, and the weighted sum feature vectors of the respective nodes can be processed through a graph neural network submodel (for example, including a plurality of fully connected layers and an activation layer, and the activation layer can be an activation layer processed using a softmax activation function) to obtain the output vectors of the respective nodes.

[0099] According to one embodiment of the present application, the first output vector of the current monitoring period is obtained according to the period encoding information and the output vectors of the respective nodes, including: combining the output vectors of the respective nodes to obtain a first output matrix; processing the period encoding information through a fourth multi-layer perception network level of the stress recognition model to obtain a period mapping vector; and multiplying the first output matrix and the period mapping vector to obtain the first output vector.

[0100] According to one embodiment of the present application, the first output matrix is composed of the output vectors of the respective nodes, and due to the attention mechanism and the connection weights, the corresponding values of the important types of physiological indicator data or HRV data in the first output matrix are higher, and the importance in the final determination of whether stress occurs is also higher.

[0101] According to one embodiment of the present application, the determination of whether stress occurs is also related to the segmented period in which the current monitoring period is located and the starting time of the monitoring period. As described above, the determination criteria and determination methods for stress in different time periods in different segmented periods are different, and therefore, when determining whether stress occurs, the feature information corresponding to the time period vector, i.e., the time period encoding information, needs to be considered. The time period encoding information can be processed by a fourth multi-layer perception network level (e.g., including multiple fully connected layers and activation layers, which can be activation layers processed using a RELU activation function) of the stress recognition model to obtain a time period mapping vector, and the time period mapping vector is multiplied by the first output matrix to fuse the feature information corresponding to the time period vector and the feature information output by each node, to obtain a first output vector, which can be used to comprehensively determine whether stress occurs in the monitoring period in combination with the features of the time period, HRV data, and multiple physiological indicator data.

[0102] According to one embodiment of the present application, according to the time period encoding information and the first output vector, the stress indicator of the patient is determined, including: determining whether the monitoring period is the first monitoring period of the segmented period in which it is located according to the time period encoding information; if the monitoring period is the first monitoring period of the segmented period in which it is located, then the first output vector is spliced with a zero vector, and the spliced vector is taken as the stress state vector of the current monitoring period, and the stress state vector is processed by a fifth multi-layer perception network level of the stress recognition model to obtain the stress indicator of the patient; if the monitoring period is not the first monitoring period of the segmented period in which it is located, then the time period encoding information is processed by a sixth multi-layer perception network level of the stress recognition model to obtain a state transition matrix; the first output vector of the previous monitoring period is processed by the state transition matrix to obtain a state transition vector; the state transition vector is spliced with the first output vector of the current monitoring period to obtain the stress state vector of the current monitoring period; and the stress state vector of the current monitoring period is processed by the fifth multi-layer perception network level of the stress recognition model to obtain the stress indicator of the patient.

[0103] According to one embodiment of the present application, if the current monitoring period is the first monitoring period in the segment period, i.e., there is no monitoring period available for reference before the current monitoring period in the segment period, the first output vector corresponding to the current monitoring period can be directly used for judgment to determine whether the patient has stress. The first output vector is spliced with a zero vector (of the same dimension as the first output vector) to obtain a stress state vector of the current monitoring period, and the stress state vector is processed through a fifth multi-layer perception network level (e.g., including multiple fully connected layers and activation layers, which can be activation layers processed using a sigmoid activation function) to obtain a stress index of the patient, which can describe the probability of the patient having stress in the current monitoring period.

[0104] According to one embodiment of the present application, if the current monitoring period is not the first monitoring period in the segment period, i.e., there is a monitoring period available for reference before the current monitoring period in the segment period, the previous monitoring period can be combined for judgment. For example, the HRV data or physiological index data in the current monitoring period and the previous monitoring period have a large change, i.e., data mutation, so the patient may have a stress reaction in the current monitoring period, and thus the first output vectors corresponding to the previous monitoring periods can be combined for judgment.

[0105] According to one embodiment of the present application, since the first output vector corresponding to the previous monitoring period describes the features of the HRV data and physiological index data in the previous monitoring period, the first output vector corresponding to the previous monitoring period can be processed through the period encoding information of the current monitoring period, so that it can be aligned with the current monitoring period in terms of features, i.e., adapted to describe whether the patient has stress in the current monitoring period. In other words, since there may be certain differences in the judgment method and standard for stress in each monitoring period, the first output vector corresponding to the previous monitoring period can be processed through the period encoding information of the current monitoring period, so that it can describe whether the patient has stress in the current monitoring period. The period encoding information can be processed through a sixth multi-layer perception network level (e.g., including multiple fully connected layers and activation layers, which can be activation layers processed using a RELU activation function) of the stress recognition model to obtain a state transition matrix, and the state transition matrix is multiplied with the first output vector of the previous monitoring period to obtain a state transition vector, i.e., the vector after the first output vector corresponding to the previous monitoring period is aligned with the current monitoring period in terms of features.

[0106] According to one embodiment of the present application, the feature information of the current monitoring period can be combined with the feature information of the previous monitoring period to determine the stress index of the patient, that is, the state transition vector is spliced with the first output vector of the current monitoring period to obtain the stress state vector of the current monitoring period, and then the stress state vector of the current monitoring period is processed through the fifth multi-layer perception network level to obtain the stress index of the patient. Thus, the features of the previous monitoring period and the features of the current monitoring period can be comprehensively judged to improve the accuracy of the judgment.

[0107] In this way, in the process of determining the stress index of the patient, the important HRV data type and the important physiological index data type can be highlighted through the attention mechanism to improve the weight, and based on the differences in the judgment criteria of each monitoring period, the attention mechanism corresponding to the monitoring period can be determined based on the time period coding information, so as to enhance the features of the important types of data, thereby improving the accuracy of determining the stress index. When determining whether the patient has stress, the segment period in which the monitoring period is located and the serial number of the monitoring period in the segment period can be combined, so that the judgment manner and the judgment criteria of different monitoring periods are considered when judging to improve the accuracy and adaptability of the judgment. In addition, the features of the previous monitoring period can be combined for judgment, and the features of the previous monitoring period can be aligned when judging, and then the features are compared comprehensively to improve the objectivity and accuracy of the comparison, thereby improving the accuracy of the stress index.

[0108] According to an embodiment of the present application, the stress recognition model can be trained before use. The training steps of the stress recognition model include: obtaining the first sample input vector, the second sample input vector and the sample period encoding information of the sample patients who appear stress in each segmented period of the perioperative period and the sample patients who do not appear stress; adjusting at least one of the preset type of HRV data and physiological index data of the sample patients who do not appear stress in each segmented period to obtain the first sample input vector, the second sample input vector and the sample period encoding information of the pseudo sample patients; selecting the sample period encoding information of the pseudo sample patients and inputting the third multi-layer perception network level of the stress recognition model to obtain the training attention mechanism; processing the training attention mechanism through the first full connection layer and the first activation layer to obtain the adjustment type recognition result; determining the attention loss function according to the adjustment type recognition result and the HRV data and physiological index data that are adjusted; selecting the first sample input vector, the second sample input vector and the sample period encoding information of any sample patient or pseudo sample patient and obtaining the corresponding training period mapping vector and training state transition matrix; processing the training state transition matrix through the second full connection layer and the second activation layer to obtain the training period vector, and splicing the training period vector with the training period mapping vector to obtain the training period recognition vector; processing the training period recognition vector through the third full connection layer and the third activation layer to obtain the training period recognition result; determining the segmented recognition loss function according to the training period recognition result, the segmented period corresponding to the sample period encoding information and the first sequence number of the start time of the monitoring period corresponding to the sample period encoding information in the segmented period; processing the first sample input vector, the second sample input vector and the sample period encoding information of the selected sample patient or pseudo sample patient through the stress recognition model to obtain the training stress index; obtaining the stress recognition loss function according to the training stress index, the first sequence number of the start time of the monitoring period corresponding to the sample period encoding information in the segmented period, the segmented period corresponding to the sample period encoding information and the label information of the sample patient or pseudo sample patient; determining the loss function of the stress recognition model according to the stress recognition loss function, the segmented recognition loss function and the attention loss function; training the stress recognition model according to the loss function of the stress recognition model to obtain the trained stress recognition model.

[0109] According to an embodiment of the present application, the way of obtaining the first sample input vector, the second sample input vector and the sample period encoding information of various sample patients is similar to the way of obtaining the first input vector, the second input vector and the period encoding information described above, which will not be repeated here.

[0110] According to one embodiment of the present application, since the number of sample patients without stress is greater than the number of sample patients with stress, at least one of the preset type of HRV data and physiological indicator data of part of the sample patients without stress in each segmented period can be manually modified, for example, the preset type of HRV data and physiological indicator data is a type that is more important for determining whether stress occurs, at least one of the preset type of HRV data and physiological indicator data can be adjusted in any monitoring period of any segmented period, so that the data characteristics of the monitoring period when stress occurs, and the monitoring period is labeled as stress, and then the first sample input vector, the second sample input vector and the sample period encoding information of the pseudo sample patient are obtained based on the similar manner as above.

[0111] According to one embodiment of the present application, as described above, the attention mechanism can enhance the HRV data and physiological indicator data of the important type (preset type), that is, increase the weight of the features corresponding to these data, so that when these data are adjusted, the stress recognition model can be trained to be sensitive to the data characteristics of stress, that is, the ability of the attention mechanism to enhance the features of the preset type of HRV data and physiological indicator data is trained. In an example, the training attention mechanism is a matrix corresponding to the monitoring period, the training attention mechanism can be obtained by a similar manner as the above attention mechanism, and it can be determined whether the training attention mechanism can enhance the features of the preset type of HRV data and physiological indicator data. During training, the sample period encoding information of the adjusted monitoring period of the pseudo sample patient can be selected for training.

[0112] According to one embodiment of the present application, the training attention mechanism can be processed through the first fully connected layer and the first activation layer (softmax activation layer) to obtain the adjustment type recognition result, that is, to determine the probability of the training attention mechanism enhancing each type of HRV data and physiological indicator data in the monitoring period, so as to determine whether the training attention mechanism can enhance the features of the preset type of HRV data and physiological indicator data based on the adjustment type recognition result.

[0113] According to one embodiment of the present application, according to the adjustment type recognition result and the adjusted HRV data and physiological indicator data, the attention loss function is determined, including: determining the attention loss function according to formula (1) , (1);

[0114] wherein, is the actual probability of the i-th type of HRV data being adjusted, is the probability of the i-th type of HRV data being adjusted according to the adjustment type recognition result, an actual probability of the jth physiological index data being adjusted, a probability of the jth physiological index data being adjusted according to the adjustment type identification result, a number of types of the HRV data, a number of types of the physiological index data, i≤ , j≤ , and i, , j, are positive integers.

[0115] According to an embodiment of the present application, in formula (1), if the ith HRV data is adjusted, that is, the ith HRV data belongs to the preset type of HRV data, and is adjusted so that the adjusted monitoring period appears the data feature of stress, then , otherwise , therefore, is , and is a cross-entropy loss function, in the training process, the cross-entropy loss function can be reduced, so that and are closer. Further, the cross-entropy loss functions corresponding to multiple types of HRV data can be summed to obtain the cross-entropy loss function of the HRV data, in the training, each and is closer.

[0116] According to an embodiment of the present application, if the jth physiological index data is adjusted, that is, the jth physiological index data belongs to the preset type of physiological index data, and is adjusted so that the adjusted monitoring period appears the data feature of stress, then , otherwise, , therefore, is , and is a cross-entropy loss function, in the training process, the cross-entropy loss function can be reduced, so that and are closer. Further, the cross-entropy loss functions corresponding to multiple types of physiological index data can be summed to obtain the cross-entropy loss function of the physiological index data, in the training, each and is closer.

[0117] According to an embodiment of the present application, the above two cross-entropy loss functions can be summed to obtain an attention loss function, in the training, the attention loss function can be reduced, so that the training attention mechanism can enhance the features of important types of HRV data and physiological index data.

[0118] In this way, the type of the HRV data and the physiological index data enhanced by the training attention mechanism can be determined through the first fully connected layer and the first activation layer, and the training attention mechanism can be enabled to enhance the features of the important type of the HRV data and the physiological index data in the training, the enhancement capability of the training attention mechanism is improved, and then the accuracy of the stress index determined by the stress recognition model is improved.

[0119] According to an embodiment of the present application, in addition to the special training of the training attention mechanism, the recognition accuracy of the stress recognition model for the segmented period to which the monitoring period belongs and the first sequence number thereof in the segmented period is also specially trained, so as to help the stress recognition model to generate accurate period encoding information based on the segmented period and the first sequence number, and then to obtain accurate attention mechanism, period mapping vector and state transition matrix, thereby helping to improve the accuracy of the stress index. The training state transition matrix can be processed through the second fully connected layer and the second activation layer (for example, the RELU activation layer), the training period vector is obtained, and the training period vector is spliced with the training period mapping vector to obtain the training period recognition vector, and then the training period recognition vector is processed through the third fully connected layer and the third activation layer to obtain the training period recognition result, that is, the segmented period and the first sequence number corresponding to the training period mapping vector and the training state transition matrix are predicted through the specially set second fully connected layer and the second activation layer and the third fully connected layer and the third activation layer, so as to determine whether the stress recognition model recognizes accurately the segmented period to which the monitoring period belongs and the first sequence number thereof in the segmented period, and the training can be performed to improve the recognition accuracy of the segmented period and the first sequence number, and help to improve the accuracy of the attention mechanism, the period mapping vector and the state transition matrix.

[0120] According to an embodiment of the present application, according to the training period recognition result, the segmented period corresponding to the sample period encoding information, and the first sequence number of the start time of the monitoring period in the segmented period, the segmented recognition loss function is determined, including: determining the segmented recognition loss function according to formula (2) ,

[0121] (2)

[0122] wherein, is the representation information of the segmented period determined according to the training period recognition result, is the representation information of the segmented period corresponding to the sample period encoding information, is the first sequence number determined according to the training period recognition result, is the first sequence number corresponding to the sample period encoding information, and is a preset weight, and .

[0123] According to one embodiment of the present application, in formula (2), The error between the segment period determined by the training period recognition result and the segment period corresponding to the sample period coding information, for example, the segment period includes three segment periods of preoperative period, intraoperative period and postoperative period, which can be represented by vectors Therefore, One of the three vectors is The representation information of the segment period output by the model is also a three-dimensional vector, The difference vector of And The modulus of The error between the real first serial number and the first serial number predicted by the model. The weighted sum of the above two errors is obtained as the segment recognition loss function, and since the importance of the segment period is higher, In the training process, the segment recognition loss function can be reduced to improve the accuracy of the segment period and the first serial number predicted by the model.

[0124] In this way, the second fully connected layer and the second activation layer and the third fully connected layer and the third activation layer can be set to predict the segment period and the first serial number corresponding to the training period mapping vector and the training state transition matrix, so as to set the segment recognition loss function to improve the accuracy of the segment period and the first serial number predicted by the model in the training process, and further improve the accuracy of the attention mechanism, the period mapping vector and the state transition matrix.

[0125] According to one embodiment of the present application, the first sample input vector, the second sample input vector and the sample period coding information of the selected sample patient or the pseudo sample patient are processed by the stress recognition model to obtain the training stress indicator, and the processing process is similar to the above process of obtaining the stress indicator, which will not be repeated here.

[0126] According to one embodiment of the present application, according to the first serial number of the start time of the monitoring period corresponding to the segment period in the segment period corresponding to the sample period coding information, the training stress indicator and the labeled information of the sample patient or the pseudo sample patient, the stress recognition loss function is obtained, including: obtaining the stress recognition loss function according to formula (3) , (3)

[0127] Among them, The probability of the occurrence of stress condition in the monitoring period is determined according to the labeled information, The training stress indicator of the monitoring period, The first serial number, The total number of cycles in the segment period to which the monitoring period belongs.

[0128] According to one embodiment of the present application, in formula (3), if the selected sample patient or pseudo sample patient has a stress reaction in the monitoring period, , otherwise, , therefore, is the cross-entropy loss function of the monitoring period corresponding to and , in the training process, the errors of and can be reduced to improve the accuracy of the stress recognition model, and can be used as the weight of the above cross-entropy loss function, wherein, in the preoperative period, the weight is , that is, the larger the first serial number of the monitoring period, the closer the monitoring period to the operation time, the greater the impact of the patient's stress reaction on the operation, and the more important the accuracy of the stress judgment, so the weight is higher, in the intraoperative period, the weight can be kept as 1, indicating that the importance of the monitoring accuracy of the stress in this period is always high, in the postoperative period, the weight is , that is, the larger the first serial number of the monitoring period, the farther the monitoring period from the operation time, the smaller the impact of the patient's stress reaction on the operation, so the weight is lower. By setting the weight, the training specificity and training efficiency can be improved.

[0129] In this way, in the training process, the weight of the training stress indicator corresponding to the cross-entropy loss function of each monitoring period can be set, and based on the influence of stress on the operation, the weight can be set according to the belonging segment period, so that the training specificity and training efficiency can be improved.

[0130] According to one embodiment of the present application, the stress recognition model can be trained based on the aforementioned stress recognition loss function, segment recognition loss function and attention loss function, for example, the three kinds of loss functions are weighted and summed, so as to improve the calculation accuracy of the model for the attention mechanism, so as to improve the enhancement ability of the model for the important HRV data and physiological index data features, and the recognition accuracy of the model for the segment period and the first serial number can be improved, so as to determine the accurate attention mechanism, period mapping vector and state transition matrix, and the recognition accuracy of the model for the stress can also be improved, that is, the performance of the stress recognition model is improved from three aspects. The parameters of the stress recognition model can be adjusted by backpropagation of the loss function, the stress recognition model is trained, and the trained stress recognition model is obtained after multiple training.

[0131] According to one embodiment of the present application, in step S8, the stress index, i.e., the probability of the patient experiencing stress in the current monitoring period, is determined based on the trained stress recognition model, and stress prompt information is determined. For example, when the stress index is higher than 0.5, stress prompt information is issued to prompt medical personnel to handle it.

[0132] The perioperative stress recognition method based on HRV intelligent monitoring according to the embodiments of the present application can obtain various signals of the patient in the perioperative period through a wearable physiological signal acquisition device, utilize the characteristics that the HRV signal can reflect the sympathetic nerve condition to recognize whether the patient experiences stress reaction in each segmented period, utilize various physiological index data for auxiliary judgment to improve the accuracy and confidence of recognition, and continuously obtain and update the collected various data to improve the timeliness of recognition. In the process of determining the stress index of the patient, the important HRV data type and the important physiological index data type can be highlighted through the attention mechanism, the weight of which is improved, and based on the differences in the judgment criteria of each monitoring period, the attention mechanism corresponding to the monitoring period can be determined based on the time period coding information, so as to enhance the features of the important data type, thereby improving the accuracy of determining the stress index. When determining whether the patient experiences stress, the segmented period in which the monitoring period is located and the serial number of the monitoring period in the segmented period are combined, so that the judgment mode and the judgment criteria of different monitoring periods are considered when judging, the accuracy and adaptability of judgment are improved, the features of the previous monitoring period can be combined for judgment, and the features of the previous monitoring period can be aligned when judging, and the features are compared after alignment, so as to improve the objectivity and accuracy of comparison, thereby improving the accuracy of the stress index. When determining the attention loss function, the HRV data and the physiological index data type enhanced by the training attention mechanism can be determined through the first fully connected layer and the first activation layer, and the training attention mechanism can be enabled to enhance the features of the important type of HRV data and physiological index data in the training, thereby improving the enhancement ability of the training attention mechanism, and further improving the accuracy of the stress recognition model in determining the stress index. When determining the segmented recognition loss function, the second fully connected layer and the second activation layer and the third fully connected layer and the third activation layer are set to predict the segmented period and the first serial number corresponding to the training period mapping vector and the training state transition matrix, so as to set the segmented recognition loss function to improve the accuracy of the predicted segmented period and the first serial number in the model in the training process, thereby improving the accuracy of the attention mechanism, the period mapping vector and the state transition matrix. When determining the stress recognition loss function, the weight of the cross-entropy loss function corresponding to the training stress index of each monitoring period can be set in the training process, the influence of stress on the operation is taken as the basis, and the weight is set according to the corresponding segmented period, so as to improve the pertinence and training efficiency of the training.

[0133] Figure 3An example of a block diagram of a perioperative stress identification system based on HRV intelligent monitoring according to an embodiment of the present application is shown, and the system comprises:

[0134] A wearing module is configured to wear a physiological signal acquisition device for a patient at the beginning of a perioperative period.

[0135] A segmented period module is configured to determine a segmented period in the perioperative period in which the current monitoring period is located.

[0136] An acquisition module is configured to determine a plurality of HRV data of the current monitoring period and physiological index data at a plurality of time points in the current monitoring period through the physiological signal acquisition device.

[0137] A first input vector module is configured to obtain a first input vector of a stress identification model according to the plurality of HRV data.

[0138] A second input vector module is configured to obtain a second input vector corresponding to each physiological index data according to the physiological index data at a plurality of time points in the current monitoring period.

[0139] A time period encoding information module is configured to determine time period encoding information according to the segmented period and the start time of the monitoring period.

[0140] A stress index module is configured to determine a stress index of the patient according to the time period encoding information, the first input vector, the second input vector and the stress identification model.

[0141] A stress prompt information module is configured to determine stress prompt information according to the stress index of the patient.

[0142] The present application can be a method, device, system and / or computer program product. The computer program product can include a computer readable storage medium having computer readable program instructions loaded thereon for performing various aspects of the present application.

[0143] Those skilled in the art should understand that the embodiments of the present application shown in the above description and the accompanying drawings are only examples and do not limit the present application. The purpose of the present application has been fully and effectively achieved. The function and structural principle of the present application has been shown and explained in the embodiments, and the embodiments of the present application can be any modification or modification without departing from the principle.

[0144] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions recorded in the above embodiments can be modified, or some or all of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A perioperative stress identification method based on HRV intelligent monitoring, characterized in that, The method comprises: when the perioperative period starts, wearing a physiological signal acquisition device for the patient; determining a segmented period in the perioperative period in which the current monitoring period is located; determining, by the physiological signal acquisition device, a plurality of HRV data of the current monitoring period and physiological index data at a plurality of time points in the current monitoring period; obtaining a first input vector of a stress recognition model according to the plurality of HRV data; obtaining a second input vector corresponding to each physiological index data according to the physiological index data at a plurality of time points in the current monitoring period; determining period encoding information according to the segmented period and the starting time point of the monitoring period; determining a stress indicator of the patient according to the period encoding information, the first input vector, the second input vector and the stress recognition model; determining stress prompt information according to the stress indicator of the patient; obtaining a first input vector of a stress recognition model according to a plurality of HRV data, comprising: grouping the plurality of HRV data into an HRV data vector; processing the HRV data vector through a first multi-layer perception network level of the stress recognition model to obtain the first input vector; obtaining a second input vector corresponding to each physiological index data according to the physiological index data at a plurality of time points in the monitoring period, comprising: grouping the physiological index data at a plurality of time points in the monitoring period into a physiological index data sequence; processing the physiological index data sequence through a 1D convolution network level of the stress recognition model to obtain the second input vector corresponding to the physiological index data; determining period encoding information according to the segmented period and the starting time point of the monitoring period, comprising: determining a first serial number of the starting time point of the monitoring period in the segmented period; combining the category information of the segmented period and the first serial number to obtain a period vector; processing the period vector through a second multi-layer perception network level of the stress recognition model to obtain the period encoding information; determining a stress indicator of the patient according to the period encoding information, the first input vector, the second input vector and the stress recognition model, comprising: taking the first input vector and the second input vector of the plurality of physiological index data as input vectors of nodes of a graph structure, respectively; processing the period encoding information through a third multi-layer perception network level of the stress recognition model to obtain an attention mechanism corresponding to the monitoring period; obtaining connection weights between nodes according to the input vectors of the nodes and the attention mechanism corresponding to the monitoring period; processing the input vectors of the nodes of the graph structure and the connection weights between the nodes through a graph neural network sub-model of the stress recognition model to obtain output vectors of the nodes; obtaining a first output vector of the current monitoring period according to the period encoding information and the output vectors of the nodes; determining a stress indicator of the patient according to the period encoding information and the first output vector; the training steps of the stress recognition model comprise: respectively obtaining first sample input vectors, second sample input vectors and sample period encoding information of sample patients who appear stress and sample patients who do not appear stress in each segmented period of the perioperative period; adjusting at least one of preset types of HRV data and physiological index data of a sample patient in each segmented period to obtain a first sample input vector, a second sample input vector and sample period encoding information of a pseudo sample patient; selecting the sample period encoding information of the pseudo sample patient and inputting the third multi-layer perception network level of the stress recognition model to obtain a training attention mechanism; processing the training attention mechanism through a first full connection layer and a first activation layer to obtain an adjustment type recognition result; determining an attention loss function according to the adjustment type recognition result and the HRV data and the physiological index data subjected to the adjustment; selecting the first sample input vector, the second sample input vector and the sample period encoding information of an arbitrary sample patient or a pseudo sample patient and obtaining a corresponding training period mapping vector and a training state transition matrix; processing the training state transition matrix through a second full connection layer and a second activation layer to obtain a training period vector, and splicing the training period vector with the training period mapping vector to obtain a training period recognition vector; processing the training period recognition vector through a third full connection layer and a third activation layer to obtain a training period recognition result; determining a segmented recognition loss function according to the training period recognition result, a segmented period corresponding to the sample period encoding information and a first sequence number of a start time of a monitoring period corresponding to the sample period encoding information in the segmented period; processing the first sample input vector, the second sample input vector and the sample period encoding information of the selected sample patient or the pseudo sample patient through the stress recognition model to obtain a training stress indicator; obtaining a stress recognition loss function according to the training stress indicator, the first sequence number of the start time of the monitoring period in the segmented period corresponding to the sample period encoding information, and the labeled information of the sample patient or the pseudo sample patient; determining a loss function of the stress recognition model according to the stress recognition loss function, the segmented recognition loss function and the attention loss function; training the stress recognition model according to the loss function of the stress recognition model to obtain a trained stress recognition model.

2. The perioperative stress identification method based on HRV intelligent monitoring according to claim 1, characterized in that, obtaining a first output vector of a current monitoring period according to the period encoding information and the output vectors of the nodes, including: combining the output vectors of the nodes to obtain a first output matrix; processing the period encoding information through a fourth multi-layer perception network level of the stress recognition model to obtain a period mapping vector; multiplying the first output matrix with the period mapping vector to obtain the first output vector.

3. The perioperative stress identification method based on HRV intelligent monitoring according to claim 1, characterized in that, determining a stress indicator of a patient according to the period encoding information and the first output vector, including: determining whether the monitoring period is the first monitoring period of a segmented period according to the period encoding information; if the monitoring period is the first monitoring period of the segmented period, splicing the first output vector with a zero vector, taking the spliced vector as a stress state vector of the current monitoring period, and processing the stress state vector through a fifth multi-layer perception network level of the stress recognition model to obtain the stress indicator of the patient; If the monitoring period is not the first monitoring period of the current segmented period, the period encoding information is processed by a sixth multi-layer perception network level of the stress recognition model to obtain a state transition matrix; The first output vector of the previous monitoring period is processed by the state transition matrix to obtain a state transition vector; The state transition vector and the first output vector of the current monitoring period are spliced to obtain a stress state vector of the current monitoring period; The stress state vector of the current monitoring period is processed by a fifth multi-layer perception network level of the stress recognition model to obtain a stress index of the patient.

4. The perioperative stress identification method based on HRV intelligent monitoring according to claim 1, characterized in that, According to the adjustment type identification result and the HRV data and physiological index data subjected to adjustment, a loss function of attention is determined, including: ， Determining an attention loss function wherein, is an actual probability of adjustment of the i-th HRV data, is a probability of adjustment of the i-th HRV data determined according to the adjustment type identification result, is an actual probability of adjustment of the j-th physiological index data, is a probability of adjustment of the j-th physiological index data determined according to the adjustment type identification result, is a number of types of the HRV data, is a number of types of the physiological index data, i≤ , j≤ , and i, , j, are positive integers.

5. The perioperative stress identification method based on HRV intelligent monitoring according to claim 1, characterized in that, According to the training period identification result, the first serial number of the start time of the monitoring period corresponding to the sample period encoding information in the segmented period, and the segmented period corresponding to the sample period encoding information, a loss function of segmentation identification is determined, including: According to the formula , determining a segment identification loss function wherein, is representation information of a segment period determined according to a training period identification result, is representation information of a segment period corresponding to sample period encoding information, is a first serial number determined according to a training period identification result, is a first serial number corresponding to sample period encoding information, and is a preset weight, and .

6. The perioperative stress identification method based on HRV intelligent monitoring according to claim 1, characterized in that, According to the first serial number of the start time of the monitoring period corresponding to the sample period encoding information in the segmented period, and the segmented period corresponding to the sample period encoding information, the stress index and the labeled information of the sample patient or the pseudo-sample patient are trained to obtain a loss function of stress identification, including: According to the formula , Stress recognition loss function is obtained wherein, is a probability of the monitoring period appearing a stress condition according to the label information, is a training stress indicator of the monitoring period, is the first serial number, is a total number of periods in a segmented period to which the monitoring period belongs.

7. A perioperative stress recognition system based on HRV intelligent monitoring, configured to perform the method according to any one of claims 1-6, characterized in that, including: The wearing module is configured to wear a physiological signal acquisition device for the patient at the beginning of the perioperative period; The segmented period module is configured to determine a segmented period in the perioperative period in which the current monitoring period is located; The acquisition module is configured to determine a plurality of HRV data of the current monitoring period and physiological index data at a plurality of time points in the current monitoring period by the physiological signal acquisition device; The first input vector module is configured to obtain a first input vector of the stress recognition model according to the plurality of HRV data; The second input vector module is configured to obtain a second input vector corresponding to each physiological index data according to the physiological index data at the plurality of time points in the current monitoring period; The period encoding information module is configured to determine period encoding information according to the segmented period and the start time of the monitoring period; The stress index module is configured to determine a stress index of the patient according to the period encoding information, the first input vector, the second input vector, and the stress recognition model; The stress prompt information module is configured to determine stress prompt information according to the stress index of the patient.

Images