Perioperative period stress identification method and system based on HRV intelligent monitoring

Abstract

The invention provides a perioperative period stress identification method and system based on HRV intelligent monitoring, and relates to the technical field of medical data processing, and the method comprises the steps: determining segmented periods in a perioperative period of a monitoring period; determining a plurality of HRV data and physiological index data, and a first input vector and a second input vector of the HRV data and the physiological index data; determining time period coding information according to the segmented time period and the starting time of the monitoring period; and determining a stress index through a stress identification model. According to the system and the method, various signals of the patient in the perioperative period can be acquired through the wearable physiological signal acquisition device, whether the patient has a stress response in each subsection period or not is identified by utilizing the characteristic that the HRV signal can reflect the sympathetic nerve condition, and various physiological index data are utilized for auxiliary judgment; according to the method, the accuracy and confidence of recognition are improved, various collected data can be continuously acquired and updated, and the timeliness of recognition is improved.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a perioperative stress identification method and system based on HRV intelligent monitoring. Background Technology

[0002] The perioperative period is a high-risk clinical phase, during which patients are prone to significant stress responses, posing challenges to preoperative preparation, intraoperative anesthesia management, and postoperative recovery. Current routine clinical monitoring methods typically include monitoring heart rate, blood pressure, and heart rate variability (HRV), but these methods suffer from problems such as response delays and low accuracy, making it difficult to reflect the state of the autonomic nervous system in a timely manner. Summary of the Invention

[0003] This invention 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 related technologies.

[0004] According to a first aspect of the present invention, a perioperative stress identification method based on HRV intelligent monitoring is provided, comprising: At the beginning of the perioperative period, patients are fitted with physiological signal acquisition devices; Determine the segmented period within the perioperative period in which the current monitoring cycle is located; Using physiological signal acquisition devices, various HRV data for the current monitoring period, as well as physiological indicator data at multiple times within the current monitoring period, are determined. Based on various HRV data, the first input vector of the stress recognition model is obtained; Based on physiological indicator data from multiple moments within the current monitoring period, a second input vector corresponding to each physiological indicator data is obtained; Based on the segmented period and the start time of the monitoring cycle, determine the time period coding information; Based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model, the patient's stress indicators are determined. Based on the patient's stress indicators, determine the stress cue information.

[0005] According to the present invention, a first input vector for a stress recognition model is obtained based on multiple HRV data, including: Combining multiple HRV data into an HRV data vector; The HRV data vector is processed through the first multilayer perceptual network layer of the stress recognition model to obtain the first input vector; Based on physiological indicator data from multiple moments within the monitoring period, a second input vector corresponding to each physiological indicator data point is obtained, including: Physiological index data from multiple moments within the monitoring period are combined into a physiological index data sequence; The physiological indicator data sequence is processed by the 1D convolutional network layer of the stress recognition model to obtain the second input vector corresponding to the physiological indicator data. Based on the segmented period and the start time of the monitoring cycle, the time period coding information is determined, including: Determine the first sequence number of the monitoring period to begin within the segmented period; The category information of the segmented period and the first sequence number are combined to obtain the time period vector; The time period vector is processed by the second multilayer perceptual network layer of the stress recognition model to obtain the time period encoding information.

[0006] According to the present invention, determining a patient's stress indicators based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model includes: The first input vector and the second input vector of various physiological index data are used as the input vectors of the nodes of the graph structure, respectively. The time period encoding information is processed by the third multi-layer perceptual network layer of the stress recognition model to obtain the attention mechanism corresponding to the monitoring period; Based on the input vector of each node and the attention mechanism corresponding to the monitoring period, the connection weights between each node are obtained; By using the graph neural network sub-model of the stress recognition model, the input vectors of the nodes in the graph structure and the connection weights between each node are processed to obtain the output vectors of each node. Based on the time period encoding information and the output vectors of each node, the first output vector of the current monitoring period is obtained; Based on the time period encoding information and the first output vector, the patient's stress indicators are determined.

[0007] According to the present invention, obtaining the first output vector of the current monitoring period based on the time period encoding information and the output vectors of each node includes: The output vectors of each node are combined to obtain the first output matrix; The time period encoding information is processed through the fourth layer of the multilayer perceptual network of the stress recognition model to obtain the time period mapping vector; The first output matrix is ​​multiplied by the time period mapping vector to obtain the first output vector.

[0008] According to the present invention, determining the patient's stress indicators based on the time period encoding information and the first output vector includes: Based on the time period coding information, determine whether the monitoring period is the first monitoring period of the segmented period; If the monitoring period is the first monitoring period of the segmented period, the first output vector is concatenated with the zero vector, and the concatenated vector is used as the stress state vector of the current monitoring period. The stress state vector is then processed by the fifth multilayer perceptual network layer of the stress recognition model to obtain the patient's stress index. If the monitoring period is not the first monitoring period of the segmented period, the time period encoding information is processed through the sixth multilayer perceptual network layer of the stress recognition model to obtain the state transition matrix. The state transition vector is obtained by processing the first output vector of the previous monitoring cycle through the state transition matrix; The state transition vector is concatenated with the first output vector of the current monitoring period to obtain the stress state vector of the current monitoring period. The stress state vector of the current monitoring period is processed by the fifth layer of the stress recognition model to obtain the patient's stress index.

[0009] According to the present invention, the training steps of the stress recognition model include: The first sample input vector, the second sample input vector, and the sample time period encoding information of the sample patients who experienced stress in each segment of the perioperative period and the sample patients who did not experience stress were obtained respectively. At least one of the preset types of HRV data and physiological index data of sample patients who did not exhibit stress was adjusted in each segment period to obtain the first sample input vector, the second sample input vector, and the sample time period encoding information of pseudo-sample patients. The sample time period encoding information of pseudo-sample patients is selected and input into the third multilayer perceptual network layer of the stress recognition model to obtain the training attention mechanism; The training attention mechanism is processed through the first fully connected layer and the first activation layer to obtain the adjusted type recognition result; Based on the adjustment type identification results and the adjusted HRV data and physiological index data, the attention loss function is determined; Select the first sample input vector, the second sample input vector, and the sample time period encoding information of any sample patient or pseudo sample patient, and obtain the corresponding training time period mapping vector and training state transition matrix; The training state transition matrix is ​​processed by the second fully connected layer and the second activation layer to obtain the training time period vector, which is then concatenated with the training time period mapping vector to obtain the training time period recognition vector. The training time period recognition vector is processed by the third fully connected layer and the third activation layer to obtain the training time period recognition result; Based on the training period identification results, the segmented periods corresponding to the sample period coding information, and the first sequence number of the start time of the monitoring period corresponding to the sample period coding information in the segmented period, the segmented identification loss function is determined. The stress recognition model processes the first sample input vector, the second sample input vector, and the sample time period encoding information of the selected sample patients or pseudo-sample patients to obtain training stress indicators. Based on the segmented period corresponding to the sample time period coding information and the first sequence number of the start time of the monitoring cycle corresponding to the sample time period coding information in the segmented period, the stress index and the annotation information of sample patients or pseudo-sample patients are trained to obtain the stress recognition loss function. Based on the stress recognition loss function, the segmentation recognition loss function, and the attention loss function, determine the loss function of the stress recognition model; Based on the loss function of the stress recognition model, the stress recognition model is trained to obtain the trained stress recognition model.

[0010] According to the present invention, an attention loss function is determined based on the adjustment type identification result and the adjusted HRV data and physiological index data, including: According to the formula ; Determine the attention loss function ,in, The actual probability of adjusting for the i-th type of HRV data. Let be the probability of adjusting the i-th type of HRV data determined based on the adjustment type identification result. The actual probability of adjusting the data for the j-th physiological indicator. The probability of adjusting the j-th physiological indicator data based on the adjustment type identification result. The number of types of HRV data. Let i be the number of types of physiological indicator data, i≤ j≤ , and i, j All are positive integers.

[0011] According to the present invention, a segmentation identification loss function is determined based on the training period identification 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 within the segmented period, including: According to the formula ; Determine the segmentation recognition loss function ,in, This is the representation information for the segmented period determined based on the recognition results of the training period. This represents the segmented period information corresponding to the coded information of the sample time period. This is the first sequence number determined based on the recognition results during the training period. This is the first sequence number corresponding to the sample time period encoding information. and As preset weights, and .

[0012] According to the present invention, based on the segmented period corresponding to the sample time period encoding information and the first sequence number of the start time of the monitoring period corresponding to the sample time period encoding information within the segmented period, stress indicators and annotation information of sample patients or dummy sample patients are trained to obtain a stress recognition loss function, including: According to the formula ; Obtain the stress recognition loss function ,in, To determine the probability of a stress condition occurring within a monitoring period based on the labeled information. To monitor training stress indicators during the monitoring period, This is the first sequence number. This represents the total number of periods within the segmented period to which the monitoring period belongs.

[0013] According to a second aspect of the present invention, a perioperative stress recognition system based on HRV intelligent monitoring is provided, comprising: Wearable module, used to equip patients with physiological signal acquisition devices at the beginning of the perioperative period; The segmented period module is used to determine the segmented period within the perioperative period in which the current monitoring cycle is located; The acquisition module is used to determine various HRV data for the current monitoring period, as well as physiological indicator data at multiple times within the current monitoring period, through a physiological signal acquisition device. The first input vector module is used to obtain the first input vector of the stress recognition model based on various HRV data. The second input vector module is used to obtain a second input vector corresponding to each physiological indicator data based on physiological indicator data at multiple times within the current monitoring period. The time period coding information module is used to determine the time period coding information based on the segmented period and the start time of the monitoring cycle; The stress index module is used to determine the patient's stress index based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model. The stress cue information module is used to determine stress cue information based on the patient's stress indicators.

[0014] By adopting the above technical solution, the present invention can achieve the following technical effects: According to the present invention, a wearable physiological signal acquisition device can acquire various signals of the patient during the perioperative period. By utilizing the characteristic that HRV signals can reflect the sympathetic nerve status, it is possible to identify whether the patient has a stress response in each segment of the period. Furthermore, multiple physiological index data can be used for auxiliary judgment to improve the accuracy and confidence of the identification. Moreover, the acquired data can be continuously acquired and updated to improve the timeliness of the identification. Furthermore, in the process of determining patient stress indicators, an attention mechanism can be used to highlight important HRV data types and important physiological indicator data types, increasing their weight. Based on the differences in judgment criteria across monitoring cycles, the attention mechanism corresponding to the monitoring cycle can be determined based on time-segment coding information, thereby enhancing the features of important data types and improving the accuracy of stress indicator determination. When determining whether a patient is experiencing stress, the segmentation period of the monitoring cycle and its sequence number within that segment can be considered to account for changes in judgment methods and criteria across different monitoring cycles, improving accuracy and adaptability. Furthermore, the system can incorporate features from previous monitoring cycles for judgment, aligning these features and performing a comprehensive comparison after feature alignment, thus improving the objectivity and accuracy of the comparison and enhancing the accuracy of stress indicators. When determining the attention loss function, the types of HRV data and physiological indicator data enhanced by the training attention mechanism can be determined by setting a first fully connected layer and a first activation layer. During training, the attention mechanism can be trained to enhance the features of important HRV data and physiological indicator data, improving its enhancement capabilities and ultimately improving the accuracy of the stress identification model in determining stress indicators. When determining the segmentation recognition loss function, the second fully connected layer, the second activation layer, the third fully connected layer, and the third activation layer can be configured to predict the segmentation period and the first index corresponding to the training time period mapping vector and the training state transition matrix. This allows for setting the segmentation recognition loss function, improving the accuracy of the model's predicted segmentation period and first index during training, thereby enhancing the accuracy of the attention mechanism, time period mapping vector, and state transition matrix. When determining the stress recognition loss function, the weights of the cross-entropy loss function corresponding to the training stress index for each monitoring time period can be set during training. Based on the impact of stress on surgery, weights are set according to the corresponding segmentation period, improving the targeting and efficiency of training. Attached Figure Description

[0015] Figure 1 A schematic flowchart of a perioperative stress identification method based on HRV intelligent monitoring according to an embodiment of the present invention is shown as an example. Figure 2 A schematic diagram illustrating the acquisition of stress indicators according to an embodiment of the present invention is shown exemplarily; Figure 3 A block diagram of a perioperative stress identification system based on HRV intelligent monitoring according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0016] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0017] Figure 1 An exemplary flowchart illustrates a perioperative stress identification method based on HRV intelligent monitoring according to an embodiment of the present invention, the method comprising: Step S1: At the beginning of the perioperative period, the patient is fitted with a physiological signal acquisition device; Step S2: Determine the segment of the perioperative period in which the current monitoring cycle is located; Step S3: Using a physiological signal acquisition device, determine various HRV data for the current monitoring period, as well as physiological indicator data for multiple moments within the current monitoring period. Step S4: Obtain the first input vector of the stress recognition model based on various HRV data; Step S5: Based on the physiological indicator data at multiple times within the current monitoring period, obtain the second input vector corresponding to each physiological indicator data. Step S6: Determine the time period coding information based on the segmented period and the start time of the monitoring cycle; Step S7: Determine the patient's stress indicators based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model; Step S8: Determine stress cue information based on the patient's stress indicators.

[0018] The perioperative stress identification method based on HRV intelligent monitoring according to an embodiment of the present invention can acquire various signals of the patient during the perioperative period through a wearable physiological signal acquisition device. By utilizing the characteristic that HRV signals can reflect the sympathetic nerve status, it can identify whether the patient has a stress response in each segment period. It can also use various physiological index data for auxiliary judgment to improve the accuracy and confidence of identification. Furthermore, it can continuously acquire and update various collected data to improve the timeliness of identification.

[0019] According to one embodiment of the present invention, in step S1, the physiological signal acquisition device can be used to acquire various HRV data, such as RMSSD data (Root Mean Square of Successive Differences), LF data (low-frequency power), HF (high-frequency power), LF / HF, etc. A set of HRV data can be obtained in each monitoring cycle. In addition, the physiological signal acquisition device can also acquire various physiological index data at multiple times in the monitoring cycle, such as electrocardiogram data, blood pressure data, blood oxygen data, body temperature data, etc. The duration of each monitoring cycle can be 10 minutes, 20 minutes, etc., and the present invention does not limit this.

[0020] According to one embodiment of the present invention, in step S2, the perioperative period can be divided into three segments: preoperative period, intraoperative period, and postoperative period. The segment in which the current monitoring cycle is located can be determined. For example, the type of segment can be manually input, or it can be determined automatically based on the surgery time. For example, the surgery time can be input as a certain time on a certain date. If the current monitoring cycle is before that date and time, then the current monitoring cycle belongs to the preoperative period. If the current monitoring cycle is after that date and time, and the patient is in the operating room as determined by the positioning device, then the current monitoring cycle belongs to the intraoperative period. If the current monitoring cycle is after the patient leaves the operating room until a set time (e.g., a certain time 7 days postoperatively), then the current monitoring cycle belongs to the postoperative period. Furthermore, when the segment period changes, the monitoring cycle can restart. For example, when the aforementioned surgery time is determined and the patient enters the operating room, even if the previous monitoring cycle (belonging to the preoperative period) has not yet ended, the previous monitoring cycle can be determined to have ended, and the first monitoring cycle of the intraoperative period can begin.

[0021] According to one embodiment of the present invention, in step S3, each monitoring cycle may include multiple time points, and various HRV data can be calculated based on electrocardiogram data and other data collected at multiple time points, thereby obtaining HRV data for the monitoring cycle.

[0022] Figure 2 A schematic diagram illustrating the acquisition of stress indicators according to an embodiment of the present invention is shown as an example.

[0023] According to one embodiment of the present invention, in step S4, the first input vector can be used to describe multiple features of HRV data. Obtaining the first input vector of the stress recognition model based on multiple HRV data includes: composing an HRV data vector from the multiple HRV data; and processing the HRV data vector through a first multilayer perceptron layer of the stress recognition model to obtain the first input vector. In the example, the first multilayer perceptron layer may include multiple fully connected layers and activation layers (e.g., layers processed using the ReLU activation function), which 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.

[0024] According to an embodiment of the present invention, in step S5, a second input vector corresponding to each physiological indicator data is obtained based on physiological indicator data from multiple times within the monitoring period. This includes: composing physiological indicator data from multiple times within the monitoring period into a physiological indicator data sequence; and processing the physiological indicator data sequence through a 1D convolutional network layer of a stress recognition model to obtain the second input vector corresponding to the physiological indicator data. The 1D convolutional network layer can process the physiological indicator data sequence. The 1D convolutional network layer may include multiple 1D convolutional layers, and each layer may include multiple convolutional kernels. After the above processing, a second input vector can be obtained, which can be used to describe various features of the physiological indicator data sequence, such as time-domain features, frequency-domain features, and variation features.

[0025] According to an embodiment of the present invention, in step S6, determining time period coding information based on the segmented period and the start time of the monitoring cycle includes: determining the first sequence number of the start time of the monitoring cycle within the segmented period; combining the category information of the segmented period and the first sequence number to obtain a time period vector; and processing the time period vector through a second multilayer perceptual network layer of a stress recognition model to obtain the time period coding information. Different segmented periods to which the current monitoring cycle belongs result in different criteria for judging stress responses. For example, in the preoperative period, patients may experience anxiety, which intensifies as the surgery time approaches, leading to abnormal electrocardiogram signals. During the intraoperative period, patients may be under anesthesia, allowing for continuous monitoring of their HRV data and multiple physiological indicators to assess their physical condition. In the postoperative period, patients may experience abnormal HRV data due to postoperative pain, which tends to subside as the wound heals and pain lessens. Therefore, the manifestation and judgment of stress response may differ at different times in different segment periods. Thus, the category information of the segment period (i.e., which can indicate which segment period the current monitoring period belongs to) and the first sequence number (i.e., each segment period may include multiple monitoring periods, and the first sequence number can indicate the sequence number of the current monitoring period among multiple monitoring periods) can be combined to obtain a time period vector. Then, the time period vector is processed through the second multi-layer perceptual network layer to obtain the time period encoding information, that is, the feature description of the time period vector, i.e., the time period encoding information, is obtained through multiple fully connected layers and activation layers.

[0026] According to one embodiment of the present invention, the first input vector represents multiple aspects of the HRV data, and the second input vector of each physiological indicator data represents multiple aspects of the physiological indicator data. The HRV data and multiple physiological indicators are interrelated and can all be used as one of the reference bases for judging stress response. Therefore, the relationship between HRV data and multiple physiological indicators can be determined by the graph neural network sub-model, and the features of HRV data can be fused with the features of multiple physiological indicator data to obtain a comprehensive feature representation and improve the accuracy of stress response judgment.

[0027] According to one embodiment of the present invention, in the graph structure, HRV data and each type of physiological indicator data can be treated as a node, with a first input vector and a second input vector as the input vectors of each node, and the connection weights between each node are determined through an attention mechanism. The higher the connection weight, the closer the relationship between the two nodes, and the higher the importance of the type of physiological indicator data or HRV data corresponding to the node for judging stress. In the example, the input vectors of two nodes can be concatenated, and the concatenated vector can be processed through an attention mechanism (e.g., the attention mechanism is a coefficient matrix) (e.g., a vector in the coefficient matrix is ​​multiplied by the concatenated vector) to obtain the connection weights between each node. Furthermore, since the judgment criteria and methods may differ for each monitoring cycle, the types of indicators monitored in each monitoring cycle may also differ. Therefore, the time-segmentation encoded information can be processed through a third multilayer perceptual network layer to obtain an attention mechanism corresponding to the monitoring cycle, thereby enhancing the type of important physiological indicator data or the features of HRV data in the monitoring cycle corresponding to the time-segmentation encoded information. In subsequent processing, each node can output comprehensive feature information that includes its own features and features from other nodes. The higher the connection weight between the node and other nodes, or the higher the weight of the node itself, the more feature information from other nodes or the node itself can be included in the comprehensive feature information. In the process of using comprehensive feature information to determine whether stress has occurred, feature information with higher weights is more important for determining whether stress has occurred. Therefore, during the training process, the parameters of the attention mechanism can be trained to determine the importance of each feature, and then determine the connection weight of the feature information of each node, so as to highlight the weight of important types of physiological indicator data or HRV data.

[0028] According to an embodiment of the present invention, after obtaining the above connection weights, the input vector of each node can be weighted and summed with the input vectors of other nodes through the connection weights to obtain the weighted summed feature vector of each node. The weighted summed feature vector of each node is then processed by a graph neural network sub-model (e.g., including multiple fully connected layers and activation layers, wherein the activation layers can be activation layers processed using the softmax activation function) to obtain the output vector of each node.

[0029] According to an embodiment of the present invention, obtaining a first output vector for the current monitoring period based on the time period coding information and the output vectors of each node includes: combining the output vectors of each node to obtain a first output matrix; processing the time period coding information through the fourth multilayer perceptual network layer of the stress recognition model to obtain a time period mapping vector; and multiplying the first output matrix with the time period mapping vector to obtain the first output vector.

[0030] According to one embodiment of the present invention, the first output matrix is ​​composed of the output vectors of each node. Due to the aforementioned attention mechanism and connection weights, the corresponding values ​​of important types of physiological indicator data or HRV data are higher in the first output matrix, and their importance in the final determination of whether stress has occurred is also higher.

[0031] According to one embodiment of the present invention, the determination of whether stress has occurred is also related to the segment period of the current monitoring cycle and the start time of the monitoring cycle. As mentioned above, different time periods within different segments have different criteria and methods for determining stress. Therefore, when determining whether stress has occurred, 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 the fourth multilayer perceptual network layer of the stress recognition model (e.g., including multiple fully connected layers and activation layers, wherein the activation layer can be an activation layer processed using the ReLU activation function) to obtain a time period mapping vector. The time period mapping vector is then multiplied by the first output matrix to fuse the feature information corresponding to the time period vector with the feature information output by each node to obtain the first output vector. This first output vector can be used to combine the features of time periods, HRV data, and features of multiple physiological indicator data to comprehensively determine whether stress has occurred in the monitoring cycle.

[0032] According to one embodiment of the present invention, determining a patient's stress index based on the time period encoding information and the first output vector includes: determining whether the monitoring period is the first monitoring period of the segmented period based on the time period encoding information; if the monitoring period is the first monitoring period of the segmented period, concatenating the first output vector with the zero vector, and using the concatenated vector as the stress state vector of the current monitoring period, and processing the stress state vector through the fifth multilayer perceptron layer of the stress recognition model to obtain the patient's stress index; if the monitoring period is not the first monitoring period of the segmented period, processing the time period encoding information through the sixth multilayer perceptron layer of the stress recognition model to obtain a state transition matrix; processing the first output vector of the previous monitoring period through the state transition matrix to obtain a state transition vector; concatenating the state transition vector with the first output vector of the current monitoring period to obtain the stress state vector of the current monitoring period; and processing the stress state vector of the current monitoring period through the fifth multilayer perceptron layer of the stress recognition model to obtain the patient's stress index.

[0033] According to one embodiment of the present invention, if the current monitoring period is the first monitoring period of a segmented period, that is, if there is no reference monitoring period before the current monitoring period in this segmented period, the first output vector corresponding to the current monitoring period can be directly used to determine whether the patient is experiencing stress. The first output vector is concatenated with a zero vector (with the same dimension as the first output vector) to obtain the stress state vector for the current monitoring period. This stress state vector is then processed by a fifth multilayer perceptron layer (e.g., including multiple fully connected layers and activation layers, where the activation layers may be activation layers processed using the sigmoid activation function) to obtain the patient's stress index, which describes the probability of the patient experiencing stress in the current monitoring period.

[0034] According to one embodiment of the present invention, if the current monitoring cycle is not the first monitoring cycle of the segmented period, that is, if there are other monitoring cycles available for reference before the current monitoring cycle within the segmented period, then the judgment can be made in conjunction with the previous monitoring cycles. For example, if there is a significant change in HRV data or physiological indicator data between the current monitoring cycle and the previous monitoring cycle, that is, a data mutation, then the patient may have a stress response in the current monitoring cycle. Therefore, the judgment can be made in conjunction with the first output vector corresponding to the previous monitoring cycle.

[0035] According to one embodiment of the present invention, since the first output vector corresponding to the previous monitoring cycle describes the features of the HRV data and physiological index data of the previous monitoring cycle, the first output vector corresponding to the previous monitoring cycle can be processed by the time period encoding information of the current monitoring cycle, thereby aligning it with the features of the current monitoring cycle. That is, making it suitable for describing whether the patient is experiencing stress in the current monitoring cycle. In other words, since the judgment method and judgment criteria for stress may differ in each monitoring cycle, the first output vector corresponding to the previous monitoring cycle can be processed by the time period encoding information of the current monitoring cycle to enable it to describe whether the patient is experiencing stress in the current monitoring cycle. The time period encoding information can be processed by the sixth multilayer perceptual network layer of the stress recognition model (e.g., including multiple fully connected layers and activation layers, wherein the activation layers can be activation layers processed using the ReLU activation function) to obtain a state transition matrix. The state transition matrix is ​​then multiplied by the first output vector of the previous monitoring cycle to obtain a state transition vector, that is, the vector after feature alignment between the first output vector corresponding to the previous monitoring cycle and the current monitoring cycle.

[0036] According to one embodiment of the present invention, the feature information of the current monitoring period can be combined with the feature information of the previous monitoring period to determine the patient's stress index. That is, the state transition vector is concatenated with the first output vector of the current monitoring period to obtain the stress state vector of the current monitoring period. Then, the stress state vector of the current monitoring period is processed through the fifth multilayer perceptual network layer to obtain the patient's stress index. Thus, a comprehensive judgment can be made by combining the features of the previous monitoring period and the features of the current monitoring period, improving the accuracy of the judgment.

[0037] In this way, during the determination of a patient's stress indicators, an attention mechanism can be used to highlight important HRV data types and important physiological indicator data types, increasing their weight. Furthermore, based on the differences in judgment criteria across monitoring cycles, the attention mechanism corresponding to the monitoring cycle can be determined based on time-segment coding information, thereby enhancing the characteristics of important data types and improving the accuracy of stress indicator determination. Additionally, when determining whether a patient is experiencing stress, the segmentation period of the monitoring cycle and its sequence number within that segment can be considered, taking into account changes in judgment methods and criteria across different monitoring cycles, thus improving the accuracy and adaptability of the judgment. Moreover, the characteristics of previous monitoring cycles can be combined for judgment, and features from previous monitoring cycles can be aligned during the judgment process. After feature alignment, a comprehensive comparison can be made, improving the objectivity and accuracy of the comparison, thereby enhancing the accuracy of stress indicators.

[0038] According to an embodiment of the present invention, the above-mentioned stress recognition model can be trained before use. The training steps of the stress recognition model include: acquiring the first sample input vector, the second sample input vector, and the sample time period encoding information of sample patients who experienced stress and sample patients who did not experience stress at various segments of the perioperative period; adjusting at least one of the preset types of HRV data and physiological index data of sample patients who did not experience stress at various segments to obtain the first sample input vector, the second sample input vector, and the sample time period encoding information of pseudo-sample patients; selecting the sample time period encoding information of pseudo-sample patients and inputting it into the third multilayer perceptron layer of the stress recognition model to obtain the training attention mechanism; processing the training attention mechanism through the first fully connected layer and the first activation layer to obtain the adjustment type recognition result; determining the attention loss function based on the adjustment type recognition result and the adjusted HRV data and physiological index data; selecting the first sample input vector, the second sample input vector, and the sample time period encoding information of any sample patient or pseudo-sample patient, and obtaining the corresponding training time period mapping vector and training state transition matrix; processing the training state transition matrix through the second fully connected layer and the second activation layer. The process involves processing the training time period vector to obtain a training time period identification vector, concatenating it with the training time period mapping vector, and then processing the training time period identification vector through a third fully connected layer and a third activation layer to obtain the training time period identification result. Based on the training time period identification result, the segmented period corresponding to the sample time period encoding information, and the first sequence number of the start time of the monitoring period corresponding to the sample time period encoding information within the segmented period, a segmented identification loss function is determined. The first sample input vector, second sample input vector, and sample time period encoding information of the selected sample patients or pseudo-sample patients are processed using a stress identification model to obtain a training stress index. Based on the segmented period corresponding to the sample time period encoding information and the first sequence number of the start time of the monitoring period corresponding to the sample time period encoding information within the segmented period, the stress index and the annotation information of the sample patients or pseudo-sample patients are trained to obtain a stress identification loss function. Based on the stress identification loss function, the segmented identification loss function, and the attention loss function, the loss function of the stress identification model is determined. Based on the loss function of the stress identification model, the stress identification model is trained to obtain the trained stress identification model.

[0039] According to one embodiment of the present invention, the method for obtaining the first sample input vector, the second sample input vector, and the sample time period encoding information of various sample patients is similar to the method for obtaining the first input vector, the second input vector, and the time period encoding information described above, and will not be repeated here.

[0040] According to one embodiment of the present invention, since the number of sample patients who did not experience stress is greater than the number of sample patients who experienced stress, at least one of the preset types of HRV data and physiological indicator data of some sample patients who did not experience stress can be manually modified in each segment period. For example, the preset types of HRV data and physiological indicator data are types that are more important for determining whether stress has occurred. At least one of the preset types of HRV data and physiological indicator data can be adjusted in any monitoring period of any segment period to make the data characteristics when stress occurs in that monitoring period, and the monitoring period is marked as stress has occurred. Then, the first sample input vector, the second sample input vector, and the sample time period encoding information of the pseudo sample patients are obtained in a similar manner as described above.

[0041] According to one embodiment of the present invention, as described above, the attention mechanism can enhance HRV data and physiological indicator data of important types (preset types), that is, increase the weight of the features corresponding to these data. Therefore, when these data are adjusted, the stress recognition model can be trained to sensitively identify the data features of stress, that is, to train the attention mechanism to enhance the features of HRV data and physiological indicator data of the preset type. In the example, the training attention mechanism is a matrix corresponding to the monitoring period. The training attention mechanism can be obtained in a manner similar to the attention mechanism described above, and it can be determined whether the training attention mechanism can enhance the features of HRV data and physiological indicator data of the preset type. During training, the sample time period encoding information of the adjusted monitoring period of spurious sample patients can be selected for training.

[0042] According to one embodiment of the present invention, the training attention mechanism can be processed by a first fully connected layer and a first activation layer (softmax activation layer) to obtain the adjustment type identification result, that is, to determine the probability that the training attention mechanism enhances each type of HRV data and physiological index data in the monitoring period, thereby judging whether the training attention mechanism can enhance the features of the preset type of HRV data and physiological index data based on the adjustment type identification result.

[0043] According to one embodiment of the present invention, determining the attention loss function based on the adjustment type identification result and the adjusted HRV data and physiological index data includes: determining the attention loss function according to formula (1). , (1); in, The actual probability of adjusting for the i-th type of HRV data. Let be the probability of adjusting the i-th type of HRV data determined based on the adjustment type identification result. The actual probability of adjusting the data for the j-th physiological indicator. The probability of adjusting the j-th physiological indicator data based on the adjustment type identification result. The number of types of HRV data. Let i be the number of types of physiological indicator data, i≤ j≤ , and i, j All are positive integers.

[0044] According to an embodiment of the present invention, in formula (1), if the i-th type of HRV data has been adjusted, that is, the i-th type of HRV data belongs to a preset type of HRV data and has been adjusted so that the adjusted monitoring period exhibits stress data characteristics, then ,otherwise ,therefore, for and The resulting cross-entropy loss function can be reduced during training, making... and Closer. Furthermore, the cross-entropy loss function corresponding to various types of HRV data can be summed to obtain the cross-entropy loss function for HRV data. During training, this can be achieved by making each type... and They are all closer.

[0045] According to one embodiment of the present invention, if the j-th physiological indicator data is adjusted, that is, the j-th physiological indicator data belongs to a preset type of physiological indicator data and is adjusted so that the adjusted monitoring period exhibits stress data characteristics, then ,otherwise, ,therefore, for and The constructed cross-entropy loss function can be reduced during training, making... and Closer. Furthermore, the cross-entropy loss functions corresponding to multiple physiological indicator data can be summed to obtain the cross-entropy loss function for the physiological indicator data. During training, this can make each... and Closer.

[0046] According to one embodiment of the present invention, the two cross-entropy loss functions can be summed to obtain the attention loss function. During training, the attention loss function can be reduced, enabling the training attention mechanism to enhance the features of important types of HRV data and physiological index data.

[0047] In this way, the types of HRV data and physiological indicators enhanced by the training attention mechanism can be determined by setting the first fully connected layer and the first activation layer. During training, the training attention mechanism can enhance the features of important types of HRV data and physiological indicators, thereby improving the enhancement capability of the training attention mechanism and thus improving the accuracy of the stress identification model in determining stress indicators.

[0048] According to one embodiment of the present invention, in addition to specifically training the attention mechanism, the stress identification model can also be specifically trained to identify the segment period to which the monitoring period belongs and the first sequence number in the segment period. This helps the stress identification model generate accurate time period coding information based on the segment period and the first sequence number, thereby obtaining accurate attention mechanism, time period mapping vector and state transition matrix, which helps to improve the accuracy of stress indicators. The training state transition matrix can be processed through a second fully connected layer and a second activation layer (e.g., a ReLU activation layer) to obtain a training time period vector. This vector is then concatenated with the training time period mapping vector to obtain a training time period identification vector. The training time period identification vector is further processed through a third fully connected layer and a third activation layer to obtain the training time period identification result. Specifically, by using specially designed second and second fully connected layers, as well as third and third fully connected layers and third activation layers, the model predicts the segmented periods and first sequence numbers corresponding to the training time period mapping vector and the training state transition matrix. This determines whether the stress identification model accurately identifies the segmented period to which the monitoring cycle belongs and its first sequence number within that segmented period. Training can then be performed to improve the accuracy of segmented periods and first sequence numbers, which helps improve the accuracy of the attention mechanism, the time period mapping vector, and the state transition matrix.

[0049] According to an embodiment of the present invention, the segmentation identification loss function is determined based on the training period identification 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 within the segmented period. This includes: determining the segmentation identification loss function according to formula (2). , (2) in, This is the representation information for the segmented period determined based on the recognition results of the training period. This represents the segmented period information corresponding to the coded information of the sample time period. This is the first sequence number determined based on the recognition results during the training period. This is the first sequence number corresponding to the sample time period encoding information. and As preset weights, and .

[0050] According to an embodiment of the present invention, in formula (2), The error between the segmented period determined by the training period identification result and the segmented period corresponding to the sample period encoding information is considered. For example, if the segmented period includes three segments: preoperative period, intraoperative period, and postoperative period, these can be represented by vectors. Therefore, Let it be one of these three vectors, and This represents the segmented time period information output by the model, and is also a three-dimensional vector. Then it is and The magnitude of the difference vector. This represents the error between the actual first sequence number and the first sequence number predicted by the model. The weighted sum of these two errors yields the segmented recognition loss function. Since the segmentation period is of high importance, therefore... During training, the segmentation recognition loss function can be reduced to improve the accuracy of the model's prediction of segment periods and first sequence numbers.

[0051] In this way, the second fully connected layer, the second activation layer, the third fully connected layer, and the third activation layer can be set to predict the segmented periods and the first sequence number corresponding to the training time period mapping vector and the training state transition matrix. This allows for the setting of a segmented recognition loss function, which improves the accuracy of the model's predicted segmented periods and the first sequence number during training, thereby enhancing the accuracy of the attention mechanism, the time period mapping vector, and the state transition matrix.

[0052] According to one embodiment of the present invention, the first sample input vector, the second sample input vector, and the sample time period encoding information of the selected sample patients or pseudo-sample patients are processed by a stress recognition model to obtain training stress indicators. The processing procedure is similar to the process of obtaining stress indicators described above, and will not be repeated here.

[0053] According to one embodiment of the present invention, based on the segmented period corresponding to the sample time period encoding information and the first sequence number of the start time of the monitoring cycle corresponding to the sample time period encoding information in the segmented period, stress indicators and annotation information of sample patients or pseudo-sample patients are trained to obtain a stress recognition loss function, including: obtaining the stress recognition loss function according to formula (3). , (3) in, To determine the probability of a stress condition occurring within a monitoring period based on the labeled information. To monitor training stress indicators during the monitoring period, This is the first sequence number. This represents the total number of periods within the segmented period to which the monitoring period belongs.

[0054] According to one embodiment of the present invention, in formula (3), if a stress response occurs during the monitoring period of the selected sample patient or spurious sample patient, then ,otherwise, ,therefore, For this monitoring period and The cross-entropy loss function can be used during training to... and To reduce errors and improve the accuracy of stress recognition models, and These can be used as weights in the aforementioned cross-entropy loss function, where, in the preoperative period, the weights are... That is, the larger the first sequence number of the monitoring period, the closer the monitoring period is to the operation time, the greater the impact of the patient's stress response on the operation, and the more important the accuracy of stress assessment. Therefore, the weight is higher. During the intraoperative period, the weight can remain at 1, indicating that the importance of stress monitoring accuracy remains high during this period. In the postoperative period, the weight is... In other words, the higher the first sequence number of the monitoring period, the further away the monitoring period is from the surgery time, and the less the patient's stress response has an impact on the surgery; therefore, the lower the weight. By setting weights, the targeting and efficiency of training can be improved.

[0055] In this way, the weights of the cross-entropy loss function corresponding to the training stress index for each monitoring period can be set during the training process. Based on the impact of stress on surgery, the weights can be set according to the segment period to which they belong, which can improve the targeting and efficiency of training.

[0056] According to one embodiment of the present invention, the stress recognition model can be trained based on the aforementioned stress recognition loss function, segmentation recognition loss function, and attention loss function. For example, the three loss functions can be weighted and summed to improve the model's accuracy in calculating the attention mechanism, thereby enhancing the model's ability to enhance features of important HRV data and physiological index data, and improving the model's accuracy in recognizing segmented periods and the first sequence number. This allows for the determination of accurate attention mechanisms, time period mapping vectors, and state transition matrices, and also improves the model's accuracy in stress recognition. In other words, the performance of the stress recognition model is comprehensively improved from three aspects. The parameters of the stress recognition model can be adjusted by backpropagating the loss functions to train the stress recognition model, and the trained stress recognition model can be obtained after multiple training iterations.

[0057] According to one embodiment of the present invention, in step S8, a stress index is determined based on the trained stress recognition model, that is, the probability that the patient will experience stress during the current monitoring period, and a stress prompt message is determined. For example, when the stress index is higher than 0.5, a stress prompt message is issued to prompt medical staff to take action.

[0058] The perioperative stress identification method based on HRV intelligent monitoring according to an embodiment of the present invention can acquire various signals of the patient during the perioperative period through a wearable physiological signal acquisition device. By utilizing the characteristic that HRV signals can reflect the sympathetic nerve status, it can identify whether the patient has a stress response in each segment period. It can also use various physiological index data for auxiliary judgment to improve the accuracy and confidence of identification. Furthermore, it can continuously acquire and update various collected data to improve the timeliness of identification. Furthermore, in the process of determining patient stress indicators, an attention mechanism can be used to highlight important HRV data types and important physiological indicator data types, increasing their weight. Based on the differences in judgment criteria across monitoring cycles, the attention mechanism corresponding to the monitoring cycle can be determined based on time-segment coding information, thereby enhancing the features of important data types and improving the accuracy of stress indicator determination. When determining whether a patient is experiencing stress, the segmentation period of the monitoring cycle and its sequence number within that segment can be considered to account for changes in judgment methods and criteria across different monitoring cycles, improving accuracy and adaptability. Furthermore, the system can incorporate features from previous monitoring cycles for judgment, aligning these features and performing a comprehensive comparison after feature alignment, thus improving the objectivity and accuracy of the comparison and enhancing the accuracy of stress indicators. When determining the attention loss function, the types of HRV data and physiological indicator data enhanced by the training attention mechanism can be determined by setting a first fully connected layer and a first activation layer. During training, the attention mechanism can be trained to enhance the features of important HRV data and physiological indicator data, improving its enhancement capabilities and ultimately improving the accuracy of the stress identification model in determining stress indicators. When determining the segmentation recognition loss function, the second fully connected layer, the second activation layer, the third fully connected layer, and the third activation layer can be configured to predict the segmentation period and the first index corresponding to the training time period mapping vector and the training state transition matrix. This allows for setting the segmentation recognition loss function, improving the accuracy of the model's predicted segmentation period and first index during training, thereby enhancing the accuracy of the attention mechanism, time period mapping vector, and state transition matrix. When determining the stress recognition loss function, the weights of the cross-entropy loss function corresponding to the training stress index for each monitoring time period can be set during training. Based on the impact of stress on surgery, weights are set according to the corresponding segmentation period, improving the targeting and efficiency of training.

[0059] Figure 3 An exemplary block diagram of a perioperative stress identification system based on HRV intelligent monitoring according to an embodiment of the present invention is shown, the system comprising: Wearable module, used to equip patients with physiological signal acquisition devices at the beginning of the perioperative period; The segmented period module is used to determine the segmented period within the perioperative period in which the current monitoring cycle is located; The acquisition module is used to determine various HRV data for the current monitoring period, as well as physiological indicator data at multiple times within the current monitoring period, through a physiological signal acquisition device. The first input vector module is used to obtain the first input vector of the stress recognition model based on various HRV data. The second input vector module is used to obtain a second input vector corresponding to each physiological indicator data based on physiological indicator data at multiple times within the current monitoring period. The time period coding information module is used to determine the time period coding information based on the segmented period and the start time of the monitoring cycle; The stress index module is used to determine the patient's stress index based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model. The stress cue information module is used to determine stress cue information based on the patient's stress indicators.

[0060] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0061] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments, and any modifications or variations of the embodiments of the present invention may be made without departing from the stated principles.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A perioperative stress identification method based on HRV intelligent monitoring, characterized in that, include: At the beginning of the perioperative period, patients are fitted with physiological signal acquisition devices; Determine the segmented period within the perioperative period in which the current monitoring cycle is located; Using physiological signal acquisition devices, various HRV data for the current monitoring period, as well as physiological indicator data at multiple times within the current monitoring period, are determined. Based on various HRV data, the first input vector of the stress recognition model is obtained; Based on physiological indicator data from multiple moments within the current monitoring period, a second input vector corresponding to each physiological indicator data is obtained; Based on the segmented period and the start time of the monitoring cycle, determine the time period coding information; Based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model, the patient's stress indicators are determined. Based on the patient's stress indicators, determine the stress cue information.

2. The perioperative stress identification method based on HRV intelligent monitoring according to claim 1, characterized in that, Based on various HRV data, the first input vector of the stress recognition model is obtained, including: Combining multiple HRV data into an HRV data vector; The HRV data vector is processed through the first multilayer perceptual network layer of the stress recognition model to obtain the first input vector; Based on physiological indicator data from multiple moments within the monitoring period, a second input vector corresponding to each physiological indicator data point is obtained, including: Physiological index data from multiple moments within the monitoring period are combined into a physiological index data sequence; The physiological indicator data sequence is processed by the 1D convolutional network layer of the stress recognition model to obtain the second input vector corresponding to the physiological indicator data. Based on the segmented period and the start time of the monitoring cycle, the time period coding information is determined, including: Determine the first sequence number of the monitoring period to begin within the segmented period; The category information of the segmented period and the first sequence number are combined to obtain the time period vector; The time period vector is processed by the second multilayer perceptual network layer of the stress recognition model to obtain the time period encoding information.

3. The perioperative stress identification method based on HRV intelligent monitoring according to claim 1, characterized in that, Based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model, the patient's stress indicators are determined, including: The first input vector and the second input vector of various physiological index data are used as the input vectors of the nodes of the graph structure, respectively. The time period encoding information is processed by the third multi-layer perceptual network layer of the stress recognition model to obtain the attention mechanism corresponding to the monitoring period; Based on the input vector of each node and the attention mechanism corresponding to the monitoring period, the connection weights between each node are obtained; By using the graph neural network sub-model of the stress recognition model, the input vectors of the nodes in the graph structure and the connection weights between each node are processed to obtain the output vectors of each node. Based on the time period encoding information and the output vectors of each node, the first output vector of the current monitoring period is obtained; Based on the time period encoding information and the first output vector, the patient's stress indicators are determined.

4. The perioperative stress identification method based on HRV intelligent monitoring according to claim 3, characterized in that, Based on the time period encoding information and the output vectors of each node, the first output vector for the current monitoring period is obtained, including: The output vectors of each node are combined to obtain the first output matrix; The time period encoding information is processed through the fourth layer of the multilayer perceptual network of the stress recognition model to obtain the time period mapping vector; The first output matrix is ​​multiplied by the time period mapping vector to obtain the first output vector.

5. The perioperative stress identification method based on HRV intelligent monitoring according to claim 3, characterized in that, Based on the time-segment encoding information and the first output vector, the patient's stress indicators are determined, including: Based on the time period coding information, determine whether the monitoring period is the first monitoring period of the segmented period; If the monitoring period is the first monitoring period of the segmented period, the first output vector is concatenated with the zero vector, and the concatenated vector is used as the stress state vector of the current monitoring period. The stress state vector is then processed by the fifth multilayer perceptual network layer of the stress recognition model to obtain the patient's stress index. If the monitoring period is not the first monitoring period of the segmented period, the time period encoding information is processed through the sixth multilayer perceptual network layer of the stress recognition model to obtain the state transition matrix. The state transition vector is obtained by processing the first output vector of the previous monitoring cycle through the state transition matrix; The state transition vector is concatenated with the first output vector of the current monitoring period to obtain the stress state vector of the current monitoring period. The stress state vector of the current monitoring period is processed by the fifth layer of the stress recognition model to obtain the patient's stress index.

6. The perioperative stress identification method based on HRV intelligent monitoring according to claim 3, characterized in that, The training steps of the stress recognition model include: The first sample input vector, the second sample input vector, and the sample time period encoding information of the sample patients who experienced stress in each segment of the perioperative period and the sample patients who did not experience stress were obtained respectively. At least one of the preset types of HRV data and physiological index data of sample patients who did not exhibit stress was adjusted in each segment period to obtain the first sample input vector, the second sample input vector, and the sample time period encoding information of pseudo-sample patients. The sample time period encoding information of pseudo-sample patients is selected and input into the third multilayer perceptual network layer of the stress recognition model to obtain the training attention mechanism; The training attention mechanism is processed through the first fully connected layer and the first activation layer to obtain the adjusted type recognition result; Based on the adjustment type identification results and the adjusted HRV data and physiological index data, the attention loss function is determined; Select the first sample input vector, the second sample input vector, and the sample time period encoding information of any sample patient or pseudo sample patient, and obtain the corresponding training time period mapping vector and training state transition matrix; The training state transition matrix is ​​processed by the second fully connected layer and the second activation layer to obtain the training time period vector, which is then concatenated with the training time period mapping vector to obtain the training time period recognition vector. The training time period recognition vector is processed by the third fully connected layer and the third activation layer to obtain the training time period recognition result; Based on the training period identification results, the segmented periods corresponding to the sample period coding information, and the first sequence number of the start time of the monitoring period corresponding to the sample period coding information in the segmented period, the segmented identification loss function is determined. The stress recognition model processes the first sample input vector, the second sample input vector, and the sample time period encoding information of the selected sample patients or pseudo-sample patients to obtain training stress indicators. Based on the segmented period corresponding to the sample time period coding information and the first sequence number of the start time of the monitoring cycle corresponding to the sample time period coding information in the segmented period, the stress index and the annotation information of sample patients or pseudo-sample patients are trained to obtain the stress recognition loss function. Based on the stress recognition loss function, the segmentation recognition loss function, and the attention loss function, determine the loss function of the stress recognition model; Based on the loss function of the stress recognition model, the stress recognition model is trained to obtain the trained stress recognition model.

7. The perioperative stress identification method based on HRV intelligent monitoring according to claim 6, characterized in that, Based on the adjustment type identification results and the adjusted HRV and physiological index data, the attention loss function is determined, including: According to the formula ; Determine the attention loss function ,in, The actual probability of adjusting for the i-th type of HRV data. Let be the probability of adjusting the i-th type of HRV data determined based on the adjustment type identification result. The actual probability of adjusting the data for the j-th physiological indicator. The probability of adjusting the j-th physiological indicator data based on the adjustment type identification result. The number of types of HRV data. Let i be the number of types of physiological indicator data, i≤ j≤ , and i, j All are positive integers.

8. The perioperative stress identification method based on HRV intelligent monitoring according to claim 6, characterized in that, Based on the training period identification results, the segmented periods 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 within the segmented period, the segmented identification loss function is determined, including: According to the formula ; Determine the segmentation recognition loss function ,in, This is the representation information for the segmented period determined based on the recognition results of the training period. This represents the segmented period information corresponding to the coded information of the sample time period. This is the first sequence number determined based on the recognition results during the training period. This is the first sequence number corresponding to the sample time period encoding information. and As preset weights, and .

9. The perioperative stress identification method based on HRV intelligent monitoring according to claim 6, characterized in that, Based on the segmented periods corresponding to the sample time period coding information and the first sequence number of the monitoring cycle start time within the segmented period, stress indicators and the annotation information of sample patients or dummy sample patients are trained to obtain a stress recognition loss function, including: According to the formula ; Obtain the stress recognition loss function ,in, To determine the probability of a stress condition occurring within a monitoring period based on the labeled information. To monitor training stress indicators during the monitoring period, This is the first sequence number. This represents the total number of periods within the segmented period to which the monitoring period belongs.

10. A perioperative stress recognition system based on HRV intelligent monitoring, used to perform the method as described in any one of claims 1-9, characterized in that, include: Wearable module, used to equip patients with physiological signal acquisition devices at the beginning of the perioperative period; The segmented period module is used to determine the segmented period within the perioperative period in which the current monitoring cycle is located; The acquisition module is used to determine various HRV data for the current monitoring period, as well as physiological indicator data at multiple times within the current monitoring period, through a physiological signal acquisition device. The first input vector module is used to obtain the first input vector of the stress recognition model based on various HRV data. The second input vector module is used to obtain a second input vector corresponding to each physiological indicator data based on physiological indicator data at multiple times within the current monitoring period. The time period coding information module is used to determine the time period coding information based on the segmented period and the start time of the monitoring cycle; The stress index module is used to determine the patient's stress index based on the time period encoding information, the first input vector, the second input vector, and the stress recognition model. The stress cue information module is used to determine stress cue information based on the patient's stress indicators.

Images