Intelligent wearable head and neck postoperative nursing monitoring system

Abstract

The application discloses an intelligent wearable head and neck postoperative nursing monitoring system and relates to the technical field of nursing monitoring.The system comprises the following steps: collecting local temperature, local pressure, micro-vibration, oral and nasal airflow sound vibration, drainage tube on-off and body position change signals output by a neck wearable monitoring patch, aligning the signals with nursing operation records, identifying nursing action events, constructing baseline, disturbance and recovery time periods, extracting local state features, and performing instability determination and risk output; the system can identify events in the local transient risk under the nursing action trigger scene, and improve the supporting capacity of the early warning result for review positioning and nursing disposal.

Description

Technical Field

[0001] This invention relates to the field of nursing monitoring technology, and more specifically, to an intelligent wearable head and neck postoperative nursing monitoring system. Background Technology

[0002] In the context of head and neck surgery nursing, current monitoring technologies generally rely on continuous vital sign monitoring, regular bedside rounds, and intermittent observation of local tissue conditions. The underlying principles are typically to detect abnormalities through indicators such as heart rate, blood oxygen saturation, respiratory rate, body temperature, skin flap color and temperature, and capillary reactivity. High-frequency monitoring is implemented within the first 24 to 72 hours post-surgery to facilitate early identification of vascular crises, airway compression, and related complications. However, truly decisive changes after head and neck surgery often do not occur during the patient's resting, stable state, but rather occur during activities such as turning, suctioning, dressing changes, attempting to swallow, getting up, adjusting pillow position, and traction. Before and after nursing procedures such as drainage tubes, local perfusion damage, increased neck compression, retention of secretions, or obstruction of drainage may rapidly cross dangerous boundaries within a short period of time. Existing technologies mostly adopt a general monitoring logic of continuous data collection, threshold comparison, and over-limit alarms. The monitoring focus is more on systemic outcome indicators, and there is a lack of specific characterization of the local, instantaneous, and chain-like deterioration process triggered by nursing actions. As a result, the system often only issues warnings when secondary abnormalities have already appeared in the systemic vital signs. It is difficult to answer in a timely manner where to focus monitoring when key nursing actions occur, within what time window to determine abnormalities, and how to distinguish between short-term recoverable fluctuations and continuous deterioration trends.

[0003] Based on this, the problem that can be identified is that existing head and neck surgery nursing monitoring technology lacks the ability to identify local instantaneous risks in scenarios triggered by nursing actions, and cannot conduct forward-looking, event-based monitoring and judgment of the rapid deterioration of local conditions before and after key nursing operations. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent wearable head and neck postoperative nursing monitoring system to address the problem that existing head and neck postoperative nursing monitoring technologies lack the ability to identify local instantaneous risks in response to nursing action triggering scenarios, and are unable to proactively and event-based monitor and determine the rapid deterioration of local conditions before and after key nursing operations such as turning over, suctioning, dressing changes, swallowing attempts, getting up, adjusting pillow position, and traction of drainage tubes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: Intelligent wearable head and neck postoperative care monitoring system, including, The data acquisition module collects local temperature signals, local pressure signals, micro-vibration signals, airflow acoustic vibration signals from the neck wearable monitoring patch of patients after head and neck surgery, drainage tube opening and closing signals, and body position change signals. It also collects nursing operation records entered by the nursing terminal and outputs raw monitoring data. The action recognition module aligns the raw monitoring data with the nursing operation records in time, identifies nursing action events corresponding to turning over, suctioning, changing dressings, attempting to swallow, getting up, adjusting pillow position, and pulling drainage tubes, and outputs a sequence of nursing action events including action type, start time, and end time. The window construction module takes each nursing action event in the nursing action event sequence as input, extracts the original monitoring data corresponding to the baseline time period before the action occurs, the disturbance time period during the action, and the recovery time period after the action ends, and outputs a monitoring window that corresponds one-to-one with each nursing action event. The status extraction module extracts features from the raw monitoring data in each monitoring window, calculates the features of neck temperature difference change, neck pressure transfer, local micro-vibration change, airflow obstruction and drainage patency, and outputs the status feature sequence corresponding to each nursing action event. The recovery judgment module compares the disturbance time period and recovery time period in each state feature sequence with the corresponding baseline time period, calculates the state deviation amplitude, recovery duration and residual deviation after recovery, and outputs the instability judgment result corresponding to each nursing action event according to the preset judgment rules. The risk output module associates the instability judgment result with the action type of the corresponding nursing action event, generates the risk level, priority review site and nursing treatment prompts corresponding to the nursing action event, and outputs the postoperative nursing warning result for head and neck surgery. By adopting the above technical solutions, a complete processing chain of acquisition, identification, windowing, extraction, judgment and output is constructed to realize continuous monitoring and event-based early warning of local state changes in the context of postoperative nursing actions in head and neck surgery, thereby improving the pertinence of abnormal identification and the timeliness of nursing response.

[0006] In a preferred embodiment, the data acquisition module is used to read the local temperature signal, local pressure signal, micro-vibration signal, airflow acoustic vibration signal from the mouth and nose, drainage tube on / off signal, and body position change signal output by the neck wearable monitoring patch, add a uniform timestamp to each signal and rearrange them according to the acquisition time sequence to output multi-channel time-series data. The amplitude continuity, rhythm continuity and channel correlation of each signal within a preset time window are calculated based on multi-channel time series data. Abnormal jump segments, missing segments and cross-channel mismatch segments are marked according to the calculation results to output time series data with quality labels. The data acquisition module is also used to time-align and link the time-series data with quality markers with the nursing operation records entered by the nursing terminal to generate raw monitoring data containing signal values, quality markers and nursing record identifiers. By adopting the above technical solutions, unified time calibration, quality marking, and nursing record linking of multi-channel signals can improve the temporal consistency of raw monitoring data, data availability, and input reliability for subsequent action recognition and status analysis.

[0007] In a preferred embodiment, the action recognition module is used to time-align the body position change signal, local pressure signal, micro-vibration signal, oral and nasal airflow acoustic vibration signal and drainage tube on / off signal in the original monitoring data with the nursing operation record, and to extract the original monitoring data of the corresponding time period using the recording time in the nursing operation record as the anchor point, and to generate nursing action candidate segments according to the direction of change, amplitude of change and duration of each signal in the corresponding time period. Based on the combined characteristics of body position change trajectory, local pressure transfer trajectory, micro-vibration rhythm change, airflow and vibration change at the mouth and nose, and drainage tube opening and closing change in each candidate nursing action segment, the nursing action type corresponding to each candidate nursing action segment is determined, and the nursing action identification result is output. The action recognition module is also used to determine the start time and end time of an action based on the start and end boundaries of each nursing action identifier on the time axis, and to associate the action type, the start time, and the end time to generate a nursing action event sequence. By adopting the above technical solutions and combining nursing operation records with multi-signal combination feature recognition of nursing action events, the accuracy of nursing action recognition and the ability to determine the start and end boundaries can be improved, thereby enhancing the pertinence of subsequent event monitoring.

[0008] In a preferred embodiment, the window construction module is used to read the action type, action start time and action end time of each nursing action event in the nursing action event sequence, and to determine the baseline time period based on the fluctuation amplitude, signal integrity and nursing record status of the original monitoring data within a continuous preset time before the action start time, and to determine the disturbance time period based on the original monitoring data between the action start time and the action end time, so as to output the baseline time period and the disturbance time period. The window construction module is also used to take the original monitoring data after the end of the action as input, calculate the deviation of local temperature signal, local pressure signal, micro-vibration signal, oral and nasal airflow acoustic vibration signal and drainage tube opening and closing signal relative to the baseline time period, and determine the continuous time period when the deviation change is lower than the preset recovery threshold as the recovery time period, so as to output a monitoring window that corresponds to each nursing action event. By adopting the above technical solution, the baseline time period, disturbance time period and recovery time period are divided around the nursing action event, and the continuous monitoring process can be transformed into an event monitoring window for a single nursing action, which makes it easier to identify short-term changes in local conditions.

[0009] In a preferred embodiment, the state extraction module is used to read local temperature signals, local pressure signals, micro-vibration signals, oral and nasal airflow acoustic vibration signals and drainage tube on / off signals in each monitoring window, and to segment them according to the baseline time period, disturbance time period and recovery time period, and to calculate the mean, rate of change, fluctuation and adjacent time difference of the corresponding signals in each time period, so as to output segmented feature data. The state extraction module is also used to calculate the temperature difference change and recovery slope of the local temperature signal between different time periods based on the segmented feature data, so as to generate neck temperature difference change characteristics; calculate the pressure offset and transfer duration of the local pressure signal between different monitoring locations, so as to generate neck pressure transfer characteristics; calculate the amplitude change and rhythm change of the micro-vibration signal between different time periods, so as to generate local micro-vibration change characteristics; calculate the ventilation amplitude change and abnormal sound vibration ratio of the oral and nasal airflow acoustic vibration signal between different time periods, so as to generate airflow obstruction characteristics; and calculate the number of on / off switching and the duration of continuous obstruction of the drainage tube on / off signal between different time periods, so as to generate drainage unobstructed characteristics. The state extraction module is also used to correlate the characteristics of neck temperature difference change, neck pressure transfer, local micro-vibration change, airflow obstruction and drainage patency in chronological order to generate a state feature sequence corresponding to each nursing action event. By adopting the above technical solution, multi-dimensional features such as temperature difference, pressure, micro-vibration, airflow and drainage can be extracted from the monitoring window, which can form a multi-angle characterization of the local condition changes after head and neck surgery, and improve the information integrity of instability analysis.

[0010] In a preferred embodiment, the state extraction module is further used to take each state feature sequence as input, and calculate the start time difference, co-variation duration and recovery order between the neck temperature difference change feature, neck pressure transfer feature, local micro-vibration change feature, airflow obstruction feature and drainage smoothness feature according to the order of appearance of each feature within the same nursing action event, so as to output feature linkage data. The state extraction module is also used to identify the transition moments of each feature from the baseline stable state to the disturbed state and from the disturbed state to the recovery state based on the feature linkage data, and to divide the continuous time period between adjacent transition moments into at least one of the pressure-dominated stage, ventilation obstruction stage, drainage abnormal stage and recovery lag stage, so as to output the state evolution fragments corresponding to the nursing action events. The state extraction module is also used to splice state evolution fragments in chronological order and combine the dominant feature type and duration of each state evolution fragment to generate an enhanced state feature sequence that characterizes the local state change process. By adopting the above technical solutions, analyzing the initiation time difference, co-variation duration and evolution stages of various state characteristics, we can reveal the linkage process and dominant change stage of local state changes, thereby enhancing the ability to characterize complex abnormal evolution processes.

[0011] In a preferred embodiment, the recovery determination module is used to read the neck temperature difference change characteristics, neck pressure transfer characteristics, local micro-vibration change characteristics, airflow obstruction characteristics, and drainage smoothness characteristics corresponding to the baseline time period, disturbance time period, and recovery time period in each state feature sequence, calculate the deviation of each feature from the baseline time period during the disturbance time period, the recovery duration when it falls back to the baseline fluctuation range during the recovery time period, and the residual deviation from the baseline time period at the end of the recovery, so as to output the sub-item determination data. The recovery judgment module is also used to determine, based on the sub-judgment data, whether the state change corresponding to each feature belongs to one of instantaneous fluctuation, delayed recovery or continuous imbalance, and to generate single-event instability judgment results corresponding to each nursing action event. By adopting the above technical solution, comparing the deviation, recovery duration, and residual deviation between the baseline time period, the disturbance time period, and the recovery time period, it is possible to distinguish between instantaneous fluctuations and continuous instability, thereby improving the accuracy of instability determination after a single nursing action.

[0012] In a preferred embodiment, the recovery determination module is further used to read the sub-determination data corresponding to the same nursing action event, calculate the number of synchronous deviations of each feature during the disturbance period, the number of synchronous lags during the recovery period, and the duration of continuous co-abnormality, so as to output the linkage determination data. The recovery judgment module is also used to correct the single event instability judgment result based on the linkage judgment data; when the number of synchronization deviations reaches the preset number threshold, the number of synchronization delays reaches the preset number threshold, or the duration of continuous co-abnormality reaches the preset duration threshold, the corresponding nursing action event is corrected to linkage instability, so as to output the enhanced instability judgment result. By adopting the above technical solutions, the synchronous deviation, synchronous lag, and continuous co-abnormalities of multiple features within the same nursing action event can be corrected in a coordinated manner, which can reduce the risk of missed judgment caused by single feature judgment and improve the ability to identify linkage instability.

[0013] In a preferred embodiment, the recovery determination module is further used to read the enhanced instability determination results corresponding to multiple consecutive nursing action events of the same patient, and calculate the number of recurrences of the same abnormal features, the transmission amplitude of the residual deviation at the end of the recovery of the previous nursing action event to the baseline time period of the next nursing action event, and the cumulative duration of continuous instability according to the order of the nursing action events on the time axis, so as to output the cumulative determination data. The recovery judgment module is also used to determine, based on the accumulated judgment data, whether the local state change corresponding to each nursing action event belongs to either single action instability or continuous action cumulative instability; when the number of repetitions reaches a preset threshold, the transmission amplitude reaches a preset transmission threshold, or the continuous instability cumulative duration reaches a preset cumulative threshold, the current nursing action event is judged as continuous action cumulative instability, and the final instability judgment result is output. By adopting the above technical solutions, analyzing the abnormal recurrence, residual transmission, and cumulative duration among multiple consecutive nursing action events, it is possible to identify continuous action instability formed across events and improve the ability to detect progressive deterioration processes.

[0014] In a preferred embodiment, the risk output module is used to read the action type and instability judgment result corresponding to each nursing action event, and to perform correlation and classification based on the action type, instability type, instability duration and instability accumulation degree to generate the risk level corresponding to each nursing action event; The risk output module is also used to determine the priority review sites and nursing treatment prompts corresponding to the current nursing action event based on the action type and the abnormal feature type corresponding to the instability judgment result, so as to output review guidance data; The risk output module is also used to combine risk level, priority review site and nursing treatment prompts to generate postoperative nursing early warning results for head and neck surgery and output them to the nursing terminal; By adopting the above technical solution, the instability judgment results are associated with the nursing action type and abnormal feature type for classification and treatment mapping, which can output early warning results for nursing execution, improve the efficiency of review and positioning and the pertinence of nursing treatment.

[0015] The technical effects and advantages of this invention are as follows: 1. Align nursing operation records with multi-channel raw monitoring data in time, construct a monitoring window around each nursing action event that includes a baseline time period, a disturbance time period, and a recovery time period, and further compare the deviation of state characteristics, recovery duration, and residual deviation after recovery within each time period. Transform the local state changes after head and neck surgery from a continuous acquisition mode to an event-based judgment mode oriented towards key nursing operations, thereby enabling more targeted and proactive identification of local instantaneous risks in nursing action trigger scenarios and improving the ability to detect rapid deterioration processes. 2. In the recovery judgment process, single-event instability judgment, multi-feature linkage correction within the same nursing action event, and abnormal repetition, residual transmission and cumulative duration analysis between multiple consecutive nursing action events are introduced in sequence. This can gradually expand from local single feature changes to linkage instability identification within the same action and cumulative instability identification across actions, thereby enhancing the ability to depict complex and progressive local deterioration processes and improving the support role of early warning results for priority review sites and nursing treatment prompts. Attached Figure Description

[0016] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Refer to the instruction manual appendix Figure 1 The intelligent wearable head and neck postoperative care monitoring system includes the following modules: The data acquisition module collects local temperature signals, local pressure signals, micro-vibration signals, airflow acoustic vibration signals from the neck wearable monitoring patch of patients after head and neck surgery, drainage tube opening and closing signals, and body position change signals. It also collects nursing operation records entered by the nursing terminal and outputs raw monitoring data. After head and neck surgery, patients' local conditions may change rapidly before and after nursing procedures such as turning over, suctioning, dressing changes, swallowing attempts, getting up, adjusting pillow position, and traction of drainage tubes. Subsequent action recognition, monitoring window construction, state extraction, and recovery determination all rely on the time consistency, data integrity, and record traceability of the front-end input data. If the acquisition times of different sensor signals are inconsistent, abnormal segments are not identified, or the nursing operation records and monitoring signals do not form a corresponding relationship, it is easy for subsequent modules to misjudge invalid fluctuations, signal gaps, or misaligned segments as real nursing actions or local abnormalities. Therefore, in this embodiment, the data acquisition module uniformly accesses the signals from each channel and sequentially completes time base unification, acquisition sequence organization, quality check, and nursing record linking, so that the output results simultaneously include the original signal values, quality markers, and nursing record identifiers. This implementation process includes the following steps: In this embodiment, the data acquisition module first reads the signals from each channel output by the wearable neck monitoring patch. Specifically, the local temperature signal is periodically output by temperature sampling units at different sampling locations, the local pressure signal is output by a pressure sampling unit, the micro-vibration signal is output by a micro-vibration sampling unit, the nasal and oral airflow acoustic vibration signal is output by an acoustic vibration sampling unit near the nasal and oral region, the drainage tube on / off signal is output by an on / off detection unit located adjacent to the drainage tube, and the body position change signal is output by a posture sampling unit. Each sampling unit provides a local acquisition time after completing one sampling cycle. This local acquisition time can be an internal timing value of the sampling unit or a clock count value allocated by the patch's main control chip. After receiving the signals from each channel, the data acquisition module first reads the system's unified clock and records the local acquisition time of each signal. The local acquisition time is converted into a unified timestamp. The specific conversion method is as follows: when the patch is powered on or the nursing shift begins, the main control chip sends a synchronization command to each sampling unit and records the unified clock value corresponding to the synchronization time. The difference between the local acquisition time of each channel signal and this synchronization time is then added to the unified clock value to obtain the corresponding unified timestamp. After the unified timestamp is appended, the data acquisition module rearranges the channel signals according to the unified timestamp and groups different channel records within the same sampling period into the same sampling group according to a preset sampling period. When a channel has multiple records within a sampling period, the record whose time is closest to the center time of that sampling period is retained. When a channel has no records within a sampling period, a null value is written to the corresponding position. Through the above processing, multi-channel time-series data is output. After outputting multi-channel time-series data, the data acquisition module performs a quality check. In this embodiment, the data acquisition module establishes a preset time window using a continuous sliding method. The preset time window can be set to 10 seconds, 20 seconds, or 30 seconds according to the ward sampling frequency and nursing rhythm. For each preset time window, the amplitude continuity, rhythm continuity, and channel correlation of each channel signal are calculated. The amplitude continuity is calculated as follows: first, the difference between two adjacent sampling times of the same channel is obtained, and then the number of times the absolute value of the difference exceeds the corresponding channel difference threshold within the preset time window is counted, which is the proportion of the total number of valid differences within the window. If the proportion exceeds the preset proportion threshold, it is determined that there is an abnormal jump trend in the channel within the window. The difference threshold is the 95th percentile of the adjacent sampling difference of the corresponding channel under historical stable wearing conditions. During the initial deployment, it can be obtained offline from the stable time period data of at least thirty head and neck surgery patients. The rhythm continuity is calculated as follows: within the preset time window, the sampling interval of the effective sampling points of the same channel is counted, and the number of times the sampling interval falls within the upper and lower fluctuation range of the target sampling period is counted, which is the proportion of the total number of sampling intervals. If the proportion is lower than the rhythm continuity threshold, it is determined that there is a missing segment or sampling interruption trend in the channel within the window; the target sampling period is given by the patch sampling configuration table, and the fluctuation range can be 10% to 20% of the target sampling period; the channel correlation is calculated as follows: within the same preset time window, select channel combinations with physiological or motor coupling relationships for comparison, such as mapping body position change signals to local pressure signals, mapping body position change signals to micro-vibration signals, and mapping oral and nasal airflow acoustic vibration signals to micro-vibration signals; for each group of channels, first determine one If a channel shows significant changes, then another channel is checked for a consistent or logically consistent response within the corresponding time offset range. If no such response is found, it is considered a cross-channel mismatch. The time offset range can be set from 1 to 3 seconds. After calculation, time periods with abnormal amplitude continuity are marked as abnormal jump segments, time periods with insufficient rhythm continuity are marked as missing segments, and time periods with unmet channel correlation are marked as cross-channel mismatch segments. If the same segment meets two or more abnormal conditions simultaneously, it is written simultaneously using a multi-label method. The final output is time series data with quality labels. After obtaining the time-series data with quality markers, the data acquisition module further aligns it with the nursing operation records entered by the nursing terminal and links them to events to generate raw monitoring data. In this embodiment, the nursing operation records entered by the nursing terminal include at least the nursing operation type, recording time, patient identifier, and recording personnel identifier. The nursing operation type is one or more of the following: turning over, suctioning, changing dressings, attempting to swallow, getting up, adjusting pillow position, and pulling drainage tubes. The recording time is the system time formed when the nursing personnel confirm the operation on the nursing terminal. The data acquisition module first reads the recording time in the nursing operation record and aligns it with the unified timestamp in the multi-channel time-series data. If there is a deviation between the nursing terminal system time and the patch's unified clock, the time conversion is performed using the time synchronization record at the beginning of the shift. Subsequently, the data acquisition module uses each nursing operation record... Centered on the recorded time, the data acquisition module extracts time-series data with quality markers within a preset attachment time range before and after the recorded time, forming candidate data segments corresponding to the nursing operation record. The preset attachment time range can be set according to the nursing operation type; for example, suctioning and turning can be set to 30 seconds before and after, dressing change can be set to 60 seconds before and after, and swallowing attempt can be set to 20 seconds before and after. For each candidate data segment, the data acquisition module attaches the corresponding nursing record identifier to all channel records within the segment. When multiple nursing operation records exist within the same time period, they are attached in chronological order of recording time, and multiple nursing record identifiers are allowed to be retained in the same channel record. After time alignment and event attachment, the data acquisition module generates raw monitoring data. Each record in the raw monitoring data includes a unified timestamp, channel identifier, signal value, quality marker, and nursing record identifier. Through the above implementation process, the data acquisition module organizes the scattered, inconsistent time-base, and event-independent patch signals and nursing records into raw monitoring data that can be directly entered into the subsequent action recognition process. This improves the temporal consistency and quality discriminability of signals from each channel, and establishes a correspondence between monitoring signals and nursing operation records, reducing the risk of misidentification caused by abnormal jumps, missing samples, and cross-channel mismatch. In practical application: For a head and neck surgery patient, the nursing staff enters nursing operation records of "turning over at 22:15" and "sputum suctioning at 22:18" on the nursing terminal. The data acquisition module first adds a unified timestamp to the signals from each channel between 22:14 and 22:19 and rearranges them. Then, it completes the quality check within a 10-second preset time window. Subsequently, it attaches quality-marked time-series data of adjacent time periods to the recording time of the two nursing records as the center, generating raw monitoring data for the subsequent action recognition module to directly identify the turning over and suctioning events.

[0019] The action recognition module aligns the raw monitoring data with the nursing operation records in time, identifies nursing action events corresponding to turning over, suctioning, changing dressings, attempting to swallow, getting up, adjusting pillow position, and pulling drainage tubes, and outputs a sequence of nursing action events including action type, start time, and end time. In postoperative nursing scenarios following head and neck surgery, changes in body position, local pressure, micro-vibration, airflow acoustic vibration, and drainage tube patency / disconnection signals typically exhibit temporally correlated changes before and after nursing actions. If judgment is based solely on the instantaneous fluctuations of a single signal, it's easy to misidentify small, spontaneous patient movements, patch loosening, or short-term noise as nursing actions, and it's also easy to break a complete nursing action into multiple isolated segments. Therefore, in this embodiment, the action recognition module uses both raw monitoring data and nursing operation records entered into the nursing terminal as input. It first extracts the raw monitoring data for the corresponding time period using the recording time in the nursing operation record as the anchor point, then extracts trajectory and change features that characterize the action's occurrence process to determine the type of nursing action, and further determines the start and end boundaries of each nursing action on the time axis. This implementation process includes the following steps: In this embodiment, the motion recognition module first reads the body position change signal, local pressure signal, micro-vibration signal, nasal and oral airflow acoustic vibration signal, and drainage tube on / off signal from the original monitoring data, and simultaneously reads the nursing operation type, recording time, patient identifier, and nursing record identifier from the nursing operation record. Specifically, the motion recognition module uses the recording time in each nursing operation record as an anchor point, extracting a first preset duration forward and a second preset duration backward to form corresponding original monitoring data segments. The first and second preset durations can be set according to the nursing operation type; for example, the extraction time before and after turning over and getting up can be set to 20 seconds and 40 seconds respectively, suctioning can be set to 10 seconds before and 30 seconds after, dressing change can be set to 30 seconds before and 60 seconds after, swallowing attempt can be set to 10 seconds before and 20 seconds after, and pillow position adjustment and drainage tube traction can be set to 15 seconds before and 25 seconds after. After extraction, the motion recognition module calculates the direction, amplitude, and duration of change of each signal within each original monitoring data segment. In the process, the direction of change is obtained by comparing the sign of the difference between the current sampling point and the previous sampling point; the amplitude of change is obtained by calculating the difference between the peak and valley values ​​within the segment; and the duration is obtained by statistically analyzing the continuous duration of segments changing in the same direction or exceeding the threshold. The threshold is determined using a channel-specific method: the threshold for changes in body position signals is the 95th percentile of the angle change under stable bed rest; the threshold for changes in local pressure signals is the 95th percentile of the pressure change under stable wearing conditions; the thresholds for changes in micro-vibration signals and oral and nasal airflow acoustic vibration signals are the 95th percentile of the amplitude change under resting conditions; and the threshold for changes in drainage tube on / off signals is the standard jump variable when switching between on / off states. The action recognition module then extracts continuous change segments that meet the corresponding threshold conditions and whose duration is greater than the minimum duration threshold as candidate segments for nursing actions. If multiple channels simultaneously meet the conditions within the corresponding time period of the same nursing operation record, continuous change segments with overlapping times or adjacent time intervals less than the merging threshold are merged to output a set of candidate segments for nursing actions. After obtaining candidate nursing action segments, the action recognition module further determines the nursing action type for each candidate segment. In this embodiment, the action recognition module first extracts combined features of body position change trajectory, local pressure transfer trajectory, micro-vibration rhythm change, airflow acoustic vibration change at the mouth and nose, and drainage tube patency / disconnection change for each candidate nursing action segment. Among them, the body position change trajectory is formed by arranging the body position change signal values ​​at continuous sampling times in chronological order; the local pressure transfer trajectory is obtained by weighted averaging of the pressure values ​​at each pressure sampling location to obtain the pressure centroid coordinates at each sampling time, and then the values ​​at each sampling location are combined... The pressure center coordinates at each moment are connected sequentially in time; the micro-vibration rhythm changes are obtained by calculating the short-time energy, dominant frequency changes, and periodic stability of the micro-vibration signals within the candidate segments; the changes in airflow acoustic vibrations at the mouth and nose are obtained by calculating the changes in ventilation amplitude, the proportion of abnormal acoustic vibration periods, and the number of acoustic vibration intervals within the candidate segments; the changes in drainage tube patency are obtained by statistically analyzing the number of state switching times and the duration of occlusion of the drainage tube patency / occlusion signals within the candidate segments; subsequently, the action recognition module matches and judges each nursing action candidate segment according to pre-established nursing action judgment rules; for example, the amplitude of the trajectory of the body position change corresponding to turning over reaches... When the turning amplitude threshold is reached and the local pressure transfer trajectory shows a continuous migration from one side to the other, the corresponding positional change trajectory for getting up is a change from a supine position to a higher angle, and the local pressure transfer trajectory migrates from the lower neck to the shoulder. During suctioning, the proportion of abnormal acoustic vibration periods in the airflow and vibration changes of the mouth and nose increases, and short-term high-frequency disturbances appear in the micro-vibration rhythm changes. During dressing changes, multiple alternating changes in the local pressure transfer trajectory and micro-vibration rhythm changes occur simultaneously, and the duration exceeds the minimum dressing change duration threshold. During attempted swallowing, a short-term interruption of ventilation followed by recovery occurs in the airflow and vibration changes of the mouth and nose, accompanied by short-term concentrated changes in the micro-vibration rhythm. (The last sentence appears to be incomplete and possibly refers to a separate event: "pillow position adjustment...") The overall positional change trajectory amplitude is lower than the turning threshold, but the local pressure transfer trajectory shows a short-range migration. The drainage tube traction corresponds to the drainage tube opening and closing change and the local pressure transfer trajectory appearing synchronously and continuously exceeding the traction recognition threshold. If a nursing action candidate segment meets two or more nursing action judgment rules at the same time, the action type with more matching features is selected first. If the number of matching features is the same, the action type consistent with the operation type registered in the nursing operation record is selected. If no specific type is registered in the nursing operation record, the action type with the highest cumulative matching score is selected to output the nursing action identification result. After obtaining the nursing action identification results, the action recognition module further determines the start and end boundaries of each nursing action on the time axis, and associates the action type, action start time, and action end time to generate a nursing action event sequence. In specific implementation, the action recognition module first takes the first effective change point of each nursing action candidate segment as the center, and traces back through consecutive preset sampling points to find the earliest moment when each combination feature transitions from a stable state to a changing state. The rules for determining the stable state are: the change amount of the body position change signal at several consecutive sampling points is lower than the body position stability threshold, the center displacement of the local pressure transfer trajectory is lower than the pressure stability threshold, the change of micro-vibration rhythm is lower than the micro-vibration stability threshold, the change of airflow acoustic vibration at the mouth and nose is lower than the airflow stability threshold, and the change of the opening and closing of the drainage tube has no state switching. When at least a preset number of the above features simultaneously meet the transition from a stable state to a changing state, the earliest moment is determined as the action start time. Subsequently, the action recognition module takes the last effective change point of the candidate segment as the center. The system searches backward through consecutive preset sampling points to find the latest moment when each combination of features returns from a changing state to a stable state and remains there. When at least a preset number of features re-satisfy the stable state determination rules and the duration reaches the stable state retention threshold, this moment is determined as the action end moment. For two nursing action candidate segments that overlap in time, the action recognition module first compares their action types and main feature sources. If they belong to the same action type and the time interval is less than the action merging threshold, they are merged into the same nursing action event, and the action start time and action end time are recalculated. If they belong to different action types, the one with the higher cumulative matching score is retained as the main event, and the one with the lower cumulative matching score is retained as a concurrent auxiliary event. Finally, the action recognition module sorts all nursing action events from morning to night according to the action start time and writes the action type, action start time, action end time, nursing record identifier, and event sequence number for each event, generating a nursing action event sequence. Through the above implementation process, the action recognition module transforms the simple temporal juxtaposition of raw monitoring data and nursing operation records into structured recognition results at the nursing action event level. This improves the ability to distinguish various nursing actions and more accurately determines the start and end boundaries of each nursing action, reducing the risk of misidentification caused by short-term noise, small-amplitude voluntary movements, or time deviations in nursing records. In practical application: the nursing staff enters the turning record at 22:15 and the suctioning record at 22:18. The action recognition module uses 22:15 and 22:18 as anchor points to extract the raw monitoring data for the corresponding time periods, identifies the corresponding candidate segments as turning and suctioning, determines the start and end times of the actions, and writes them into the nursing action event sequence in chronological order.

[0020] The window construction module takes each nursing action event in the nursing action event sequence as input, extracts the original monitoring data corresponding to the baseline time period before the action occurs, the disturbance time period during the action, and the recovery time period after the action ends, and outputs a monitoring window that corresponds one-to-one with each nursing action event. After head and neck surgery, the local condition of patients typically undergoes a continuous process of change before and after nursing actions, from relative stability to disturbance by the action, and then to gradual recovery. If a fixed length of data is directly extracted based solely on the moment of the action recording, it is easy to mistake pre-existing abnormal fluctuations for normal baselines, and to mistake persistent abnormalities that have not yet recovered after the action for short-term disturbances. Therefore, in this embodiment, the window construction module takes the sequence of nursing action events as input, first determines the baseline time period before the action and the disturbance time period during the action, and then identifies the recovery time period by combining the original monitoring data after the action ends, thereby organizing the continuous time series data into monitoring windows that correspond one-to-one with individual nursing action events. This implementation process includes the following steps: In this embodiment, the window construction module first reads the action type, start time, and end time of each nursing action event in the nursing action event sequence, and retrieves local temperature signals, local pressure signals, micro-vibration signals, nasal and oral airflow acoustic vibration signals, and drainage tube patency / disconnection signals from the original monitoring data before and after the corresponding time period of the nursing action event. Subsequently, the window construction module uses the original monitoring data within a continuously preset duration before the start time of the action as the candidate baseline interval. The continuously preset duration is determined according to the nursing action type and the ward's historical nursing data. For example, the baseline observation time before turning over, getting up, and adjusting the pillow position can be set to 20 seconds. The baseline observation time before suctioning and attempted swallowing can be set to 10 to 20 seconds, and the baseline observation time before dressing changes and drainage tube traction can be set to 30 to 60 seconds. The window construction module then calculates the fluctuation amplitude and signal integrity of each channel signal within the candidate baseline interval. The fluctuation amplitude is represented by the difference between the maximum and minimum values ​​of each channel signal within the candidate baseline interval, divided by the baseline amplitude of that channel during a stable wearing period to obtain the normalized fluctuation amplitude. Signal integrity is obtained by statistically analyzing the proportion of effective sampling points to the theoretical number of sampling points within the candidate baseline interval. Effective sampling points are those not marked as abnormal by quality. The sampling points for common jump segments, missing segments, or cross-channel mismatch segments; the window construction module further reads the nursing record status corresponding to the candidate baseline interval in the nursing operation record, and determines whether there are other nursing operation interferences in the candidate baseline interval; if there is a nursing record identifier different from the current nursing action event in the candidate baseline interval, or there are "processing", "incomplete", or "repeated operation" records registered by the nursing terminal, then the candidate baseline interval is determined to be subject to nursing interference; if there are no other nursing operation interferences, and the normalized fluctuation amplitude of each channel is lower than the corresponding fluctuation threshold and the signal integrity is higher than the integrity threshold, then the candidate baseline interval is... The baseline interval is selected as the baseline time period. When the most recent candidate baseline interval does not meet the conditions, the window construction module slides forward to search for new candidate baseline intervals according to the preset backtracking step size. If no interval that meets the conditions is found within the maximum backtracking range, the candidate baseline interval with the highest signal integrity and the smallest normalized fluctuation amplitude is selected as the alternative baseline time period, and the "baseline confidence reduced" label is written in the monitoring window corresponding to the nursing action event. After the baseline time period is determined, the window construction module directly determines the original monitoring data between the start time and the end time of the action as the disturbance time period, and outputs the baseline time period and the disturbance time period. After obtaining the baseline time period and the disturbance time period, the window construction module further reads the raw monitoring data after the action end time to identify the recovery time period. Specifically, the window construction module establishes a recovery observation interval starting from the action end time and proceeds backward. Within this recovery observation interval, it calculates the deviation changes of the local temperature signal, local pressure signal, micro-vibration signal, nasal and oral airflow acoustic vibration signal, and drainage tube on / off signal relative to the baseline time period. For the local temperature signal, the deviation change is taken as the absolute value of the difference between the temperature value at each sampling time within the recovery observation interval and the average temperature value of the baseline time period. For the local pressure signal, the deviation change is... The distance between the pressure centroid coordinates at each sampling time and the average pressure centroid coordinates of the baseline time period is taken, or the absolute value of the difference between the pressure value and the average pressure value of the baseline time period is taken. For micro-vibration signals and oral-nasal airflow acoustic vibration signals, the deviation change is taken as the absolute value of the difference between the short-time energy, dominant frequency, or ventilation amplitude and the mean of the baseline time period, respectively. For drainage tube on / off signals, the deviation change is taken as the number of inconsistencies between the current on / off state and the main state of the baseline time period or the duration of continuous blockage. The window construction module then normalizes the deviation changes of each channel, and the normalization benchmark is the mean, standard deviation, or stable value of the corresponding channel within the baseline time period. The upper bound of the fluctuation is defined. After normalization, the window construction module sums the normalized deviation changes of multiple channels according to preset weights to obtain the comprehensive deviation value at each sampling time. The preset weights are determined according to the type of nursing action. For example, in the turning event, the local pressure signal and the signal related to the change in body position have higher weights; in the suction event, the airflow sound vibration signal and the micro-vibration signal of the mouth and nose have higher weights; and in the dressing change and drainage tube traction event, the local pressure signal and the opening and closing signal of the drainage tube have higher weights. Subsequently, the window construction module defines the continuous period when the comprehensive deviation value is lower than the preset recovery threshold as the recovery period. The preset recovery threshold can be taken as the normal value. The 95th percentile of the overall deviation value in the recovery sample is used; to avoid misjudgment caused by short-term accidental drops, the length of the continuous time period is required to be no less than the minimum recovery time threshold; when no continuous time period meeting the conditions appears within the maximum recovery observation range after the action ends, the window construction module marks the data between the end time of the action and the end time of the maximum recovery observation range as an unrecovered interval, and outputs this unrecovered interval as the recovery time period, while writing the "Recovery incomplete" label; finally, the window construction module associates the baseline time period, the disturbance time period, and the recovery time period to generate a monitoring window that corresponds one-to-one with each nursing action event; Through the above implementation process, the window construction module organizes the continuous raw monitoring data corresponding to nursing action events into baseline time periods, disturbance time periods, and recovery time periods with clear temporal semantics. This provides a stable comparison benchmark and clear analysis boundaries for the state extraction module, and facilitates the subsequent recovery judgment module in calculating the state deviation magnitude, recovery duration, and residual deviation after recovery. In practical applications: for a nursing action event of "turning over at 22:15 and ending at 22:15:18," the window construction module first reads the raw monitoring data from 22:14:40 to 22:15 and compares multiple... Based on the normalized fluctuation amplitude, signal integrity, and nursing operation record status of the candidate intervals, the period from 22:14:42 to 22:15 was ultimately determined as the baseline time period, and the period from 22:15 to 22:15:18 was determined as the disturbance time period. Subsequently, the original monitoring data after 22:15:18 was read, and the deviation of each channel relative to the baseline time period was calculated point by point. The comprehensive deviation value was obtained by combining the weights corresponding to the turning events. When the comprehensive deviation value was lower than the preset recovery threshold for 8 consecutive seconds starting from 22:15:26, the period from 22:15:26 to 22:15:34 was determined as the recovery time period.

[0021] The status extraction module extracts features from the raw monitoring data in each monitoring window, calculates the features of neck temperature difference change, neck pressure transfer, local micro-vibration change, airflow obstruction and drainage patency, and outputs the status feature sequence corresponding to each nursing action event. Before and after postoperative nursing actions in head and neck surgery, local temperature, local pressure, micro-vibration, nasal and oral ventilation, and drainage status typically evolve continuously along a process of "single feature initiation—multiple feature linkage—partial feature recovery first, partial feature recovery later." If only a single signal is compared instantaneously, or only separate individual features are output, it is difficult to reflect the dominant factors, linkage relationships, and evolutionary sequence of local state changes within the same nursing action event. Therefore, in this embodiment, the state extraction module first analyzes the local temperature signal, local pressure signal, micro-vibration signal, nasal and oral airflow acoustic vibration signal, and drainage tube status within each monitoring window. The interrupted signal is segmented according to the baseline time period, disturbance time period, and recovery time period, and segmented characteristic data within each time period are calculated. Based on the segmented characteristic data, characteristics of neck temperature difference change, neck pressure transfer, local micro-vibration change, airflow obstruction, and drainage patency are generated, and these characteristics are correlated chronologically to form a state characteristic sequence. Furthermore, the onset time difference, co-variation duration, recovery sequence, and state transition process between each characteristic are analyzed to divide the state evolution segments within nursing action events and generate an enhanced state characteristic sequence. This implementation process includes the following steps: In this embodiment, the state extraction module first reads the monitoring window corresponding to a single nursing action event, and extracts the sampling data of local temperature signal, local pressure signal, micro-vibration signal, nasal and oral airflow acoustic vibration signal, and drainage tube on / off signal from the monitoring window within the baseline time period, disturbance time period, and recovery time period. To ensure consistency in subsequent calculations, the state extraction module first performs unified preprocessing on the signals of each channel within each time period: for local temperature signals, they are directly sorted according to a unified timestamp before being included in the calculation; for local pressure signals, multi-position pressure sequences are formed according to the pressure sampling positions on the neck wearable monitoring patch, and different monitoring positions are multiple pressure sampling positions pre-arranged on the neck wearable monitoring patch, each pressure sampling position having a fixed relative coordinate when the patch is installed; for micro-vibration signals and nasal and oral airflow acoustic vibration signals, short-time framing is performed according to a fixed-length analysis window, the analysis window length of which can be 0.5 seconds or 1 second. Or 2 seconds; for the opening and closing signal of the drainage tube, the original opening and closing state sequence is maintained, and the state value at each sampling moment is uniformly represented as an open state or a closed state; after preprocessing, the state extraction module calculates the mean, rate of change, volatility, and difference between adjacent moments of the signal of each channel in each time period to output segmented feature data; among them, the mean is the arithmetic mean of all valid sampled values ​​in the time period, the rate of change is the difference between the signal value at the end of the time period and the signal value at the beginning of the time period divided by the duration of the time period, the volatility is the standard deviation or root mean square value of the signal value in the time period relative to the mean of the time period, and the difference between adjacent moments is the difference between each sampling moment and the previous sampling moment; if the number of valid sampling points in a certain time period is less than the preset minimum number of sampling points, the state extraction module writes an "insufficient sampling" flag for the time period and reduces the confidence level of the corresponding feature of the time period to a preset low confidence level; after the above calculation, the state extraction module outputs segmented feature data; After obtaining the segmented feature data, the state extraction module generates five basic state features. For the neck temperature difference change feature, the state extraction module first calculates the difference between the average temperature value of the local temperature signal during the disturbance period and the average temperature value during the baseline period to obtain the disturbance temperature difference change. Then, it calculates the difference between the average temperature value at the end of the recovery period and the average temperature value during the baseline period to obtain the temperature difference residual at the end of the recovery period. Simultaneously, using the absolute deviation of the temperature value at each sampling time within the recovery period relative to the average temperature value of the baseline period as the ordinate and time as the abscissa, it uses least squares linear fitting to obtain the recovery slope. For the neck pressure transfer feature, the state extraction module calculates the pressure centroid coordinates at each sampling time based on the pressure value at each pressure sampling location and the corresponding fixed relative coordinates. Then, it calculates the displacement distance and displacement direction of the pressure centroid relative to the average pressure centroid of the baseline period within the disturbance period, and counts the duration for which the displacement is continuously higher than the displacement threshold to obtain the pressure offset and transfer duration. For the local micro-vibration change feature... The state extraction module calculates the mean amplitude, dominant frequency, and dominant frequency difference between adjacent analysis windows of the micro-vibration signal within each analysis window. It then compares the mean amplitude difference and dominant frequency difference between the disturbance time period and the baseline time period, and between the recovery time period and the baseline time period, to obtain the amplitude change and rhythm change. For airflow obstruction characteristics, the state extraction module first calculates the ventilation amplitude, energy concentration frequency band, and short-term zero-crossing rate within each analysis window of the oral and nasal airflow acoustic vibration signal. Then, it identifies abnormal acoustic vibration periods according to the abnormal acoustic vibration judgment rules. Subsequently, the state extraction module calculates the change in ventilation amplitude between the disturbance time period and the baseline time period, and uses the proportion of abnormal acoustic vibration analysis windows to the total number of analysis windows in that time period as the proportion of abnormal acoustic vibration periods, thereby generating airflow obstruction characteristics. For unobstructed drainage characteristics, the state extraction module counts the number of state switching times of the drainage tube on / off signal within the disturbance time period and the recovery time period, and calculates the longest and cumulative duration of continuous obstruction. After completing the above five types of calculations, the state extraction module outputs five types of basic state characteristics. After obtaining the five basic state features, the state extraction module further correlates them according to time sequence to generate a state feature sequence corresponding to each nursing action event. In specific implementation, the state extraction module first determines the start time, peak time, and fall time for each basic state feature. The start time refers to the earliest time when the deviation value corresponding to the feature first continuously exceeds the feature's start time threshold; the peak time refers to the time when the deviation value corresponding to the feature reaches its maximum value; and the fall time refers to the earliest time when the deviation value corresponding to the feature falls below the feature's recovery threshold again and remains below it for at least a preset duration. The state extraction module writes the five basic state features and their corresponding feature values ​​into the same timeline according to the above three times and forms a feature entry sequence according to time sequence. Each feature entry includes at least the feature type, start time, peak time, fall time, peak size, duration, and reliability. The final state feature sequence is used to characterize the basic sequential relationship and the strength of individual changes of various local state changes within the same nursing action event. After generating the state feature sequence, the state extraction module further uses each state feature sequence as input to calculate the start-up time difference, co-variation duration, and recovery order among the features within the same nursing action event, and outputs feature linkage data. In specific implementation, the start-up time difference is obtained by the difference between the start times of any two types of features, the co-variation duration is obtained by statistically analyzing the overlap length when the deviation values ​​of two types of features are simultaneously higher than their respective start-up thresholds, and the recovery order is obtained by comparing the order in which each feature falls back. The state extraction module combines the five basic state features in pairs to obtain the above three types of linkage quantities and forms a feature linkage matrix. When the co-variation duration of two types of features is lower than the minimum co-variation duration threshold, the co-variation relationship is marked as a weak linkage relationship. Thus, the state extraction module outputs feature linkage data. After obtaining the feature linkage data, the state extraction module further identifies the transition moments of each feature from the baseline stable state to the perturbed state and from the perturbed state to the recovery state. It then divides the continuous time interval between adjacent transition moments into at least one of the following: pressure-dominated stage, ventilation obstruction stage, drainage abnormality stage, and recovery lag stage, to output the state evolution fragment corresponding to the nursing action event. Specifically, the state extraction module first merges the start and fall moments of each feature into a set of transition moments in chronological order. Then, it selects the continuous time interval between any two adjacent transition moments as candidate evolution periods and calculates the average deviation level and sustained activity level of each feature within these candidate evolution periods. If, within a certain candidate evolution period, the neck is pressure-dominated... If the average deviation level of the neck compression transfer feature is the largest and exceeds the pressure-dominated threshold, the candidate evolution period is classified as the pressure-dominated stage. If the average deviation level of the airflow obstruction feature is the largest and the proportion of abnormal sound vibration periods is higher than the ventilation obstruction threshold, it is classified as the ventilation obstruction stage. If the proportion of continuous obstruction duration in the drainage unobstructed feature is higher than the drainage abnormality threshold, it is classified as the drainage abnormality stage. If at least one of the neck compression transfer feature, local micro-vibration change feature, or local temperature difference change feature has decreased, but the airflow obstruction feature or drainage unobstructed feature has not decreased, and this non-declined state continues to exceed the recovery lag threshold, it is classified as the recovery lag stage. Finally, the state extraction module outputs the state evolution results of nursing action events consisting of multiple state evolution segments. After obtaining the state evolution segments, the state extraction module further splices the state evolution segments in chronological order and combines the dominant feature type and duration of each state evolution segment to generate an enhanced state feature sequence. In specific implementation, the state extraction module sorts the state evolution segments from early to late according to their start time, and then merges two adjacent state evolution segments with the same stage type and a time interval below the segment merging threshold into the same long segment. Subsequently, the state extraction module generates a segment identifier for each state evolution segment, which includes at least the stage type, dominant feature type, segment duration, and transition relationship with the previous segment. The state extraction module then associates each segment identifier with the feature entry sequence in the original state feature sequence, so that each feature entry not only retains its own start time, peak time, and fall time, but also retains the stage type of the state evolution segment it belongs to, the dominant feature type of the stage, and the duration of the stage. The final generated enhanced state feature sequence contains both basic change information of individual state features and information on multi-feature linkage and state evolution stages, which can be directly called by the subsequent recovery judgment module. Through the above implementation process, the state extraction module can not only generate five basic state features from the monitoring window, but also further reveal the temporal linkage relationship, dominant change stage, and state evolution process of each feature. This expands the local state changes within the same nursing action event from single feature comparison to process-oriented, linkage-oriented, and phased representation. In practical applications: For a turning nursing action event of a head and neck postoperative patient, the state extraction module first reads the sampling data of each channel signal from the corresponding monitoring window during the baseline time period, disturbance time period, and recovery time period, and calculates the segmented feature data. Subsequently, it generates five basic state features and writes the above features into the state feature sequence. Then, it further calculates the initiation time difference, covariation duration, and recovery sequence between the neck pressure transfer feature and the airflow obstruction feature, identifying that after turning, the patient first enters the pressure-dominated stage, then the airflow obstruction stage, and finally the recovery lag stage. The above stages are spliced ​​together in chronological order to generate the enhanced state feature sequence corresponding to the turning nursing action event.

[0022] The recovery judgment module compares the disturbance time period and recovery time period in each state feature sequence with the corresponding baseline time period, calculates the state deviation amplitude, recovery duration and residual deviation after recovery, and outputs the instability judgment result corresponding to each nursing action event according to the preset judgment rules. In postoperative nursing scenarios following head and neck surgery, local state changes after a single nursing action are not always manifested as a clear exceedance of a single feature. More often, multiple features deviate or lag synchronously within the same nursing action event, or gradually accumulate between consecutive nursing action events through residual deviation transmission and abnormal repetition. If judgment is based solely on the degree of abnormality of a single feature at a single moment, recoverable short-term fluctuations are easily misjudged as risks, and the continuous deterioration process formed by the linkage of multiple features or cross-action accumulation is easily overlooked. Therefore, in this embodiment, the recovery judgment module first performs item-by-item judgment on the sequence of state features within the same nursing action event, distinguishing between instantaneous fluctuations, delayed recovery, and continuous instability. Then, it analyzes the synchronous deviation, synchronous lag, and continuous co-abnormal relationships of each feature within the same nursing action event, and performs linkage correction on the single-event instability judgment result. Finally, it analyzes the number of repetitions of the same abnormal feature, the transmission amplitude of residual deviation from the previous nursing action event to the baseline time of the next nursing action event, and the cumulative duration of continuous instability between consecutive nursing action events for the same patient, in order to identify continuous action cumulative instability. This implementation process includes the following steps: In this embodiment, the recovery determination module first reads the state feature sequence corresponding to a single nursing action event. The state feature sequence includes at least the neck temperature difference change feature, neck pressure transfer feature, local micro-vibration change feature, airflow obstruction feature, and drainage unobstructed feature corresponding to the baseline time period, disturbance time period, and recovery time period. Each type of feature includes a deviation value, start time, peak time, fall time, duration, and reliability. Subsequently, the recovery determination module calculates the deviation of each feature from the baseline time period during the disturbance time period, the recovery duration when it falls back to the baseline fluctuation range during the recovery time period, and the residual value relative to the baseline time period at the end of the recovery. The residual deviation is determined by outputting sub-item judgment data. The deviation magnitude is the absolute value of the difference between the peak deviation of the feature during the disturbance period and the mean value of the corresponding feature during the baseline period, further normalized using the upper bound of the stable fluctuation of the feature during the baseline period to obtain the normalized deviation magnitude. The recovery duration is the time length from the start of the recovery period to the moment when the feature falls back to the baseline fluctuation range and remains continuously at or above the minimum holding time threshold. The residual deviation is the absolute value of the difference between the feature deviation value at the end of the recovery period and the mean value of the baseline period, also normalized using the upper bound of the stable fluctuation. For features that fail to fall back to the baseline fluctuation range within the recovery period... For features within the specified range, the recovery duration is directly recorded as the total recovery time period and marked as "no decline". After completing the above calculation, the recovery determination module classifies and determines the state changes of each type of feature based on the sub-determination data. The specific determination rules are as follows: when the normalized deviation amplitude reaches below the first deviation threshold, and the recovery duration does not exceed the first recovery threshold, and the residual deviation is lower than the first residual threshold, it is determined to be an instantaneous fluctuation; when the normalized deviation amplitude reaches above the first deviation threshold but is lower than the second deviation threshold, or the recovery duration exceeds the first recovery threshold but does not exceed the second recovery threshold, and the residual deviation is higher than the first residual threshold but lower than the second residual threshold, it is determined to be an instantaneous fluctuation. If the normalized deviation reaches or exceeds the second deviation threshold, or the recovery duration exceeds the second recovery threshold, or the residual deviation reaches or exceeds the second residual threshold, it is determined to be a persistent instability. The recovery determination module then integrates the classification results of the five types of features. If all features are determined to be instantaneous fluctuations, the nursing action event is determined to be a single-event stable fluctuation. If at least one feature is determined to be delayed recovery and no feature is determined to be persistent instability, the nursing action event is determined to be a single-event delayed recovery. If at least one feature is determined to be persistent instability, the nursing action event is determined to be a single-event instability, and the sub-determination data and single-event instability determination results are output. After obtaining the single-event instability judgment result, the recovery judgment module further reads all the component judgment data corresponding to the same nursing action event, calculates the number of synchronous deviations of each feature during the disturbance period, the number of synchronous lags during the recovery period, and the duration of continuous co-abnormality, and outputs the linkage judgment data. In specific implementation, the recovery judgment module first determines the deviation activity interval and lag activity interval for each type of feature. The deviation activity interval refers to the time interval during which the feature deviation value is continuously higher than the corresponding initiation threshold, and the lag activity interval refers to the time interval during which the feature deviation value is still continuously higher than the corresponding recovery threshold during the recovery period. Subsequently, each sampling time is traversed point by point during the disturbance period, and the number of features that are simultaneously in the deviation activity interval at the same sampling time is counted, and the maximum value is taken as the number of synchronous deviations. Similarly, the number of features that are simultaneously in the lag activity interval is counted point by point during the recovery period, and the maximum value is taken as the number of synchronous lags. The calculation method for the duration of continuous co-abnormality is as follows: the overlapping time of any two or more types of features that are simultaneously in the deviation activity interval or lag activity interval is merged, and the length of the longest continuous overlapping time after merging is counted as the duration of continuous co-abnormality. To avoid occasional However, short-term overlap can cause misjudgment. Continuous co-abnormal duration is only counted when the overlap duration reaches the minimum co-abnormal duration threshold. After completing the linkage calculation, the recovery judgment module corrects the single-event instability judgment result based on the linkage judgment data. The specific correction rules are as follows: when the number of synchronization deviations reaches a preset threshold, it indicates that there are at least a preset number of characteristics with significant deviations during the nursing action disturbance phase. The recovery judgment module corrects the current nursing action event from single-event stable fluctuation or single-event delayed recovery to linkage instability. When the number of synchronization lags reaches a preset threshold, it indicates that there are still at least a preset number of characteristics with significant deviations during the recovery phase. If the synchronization of a few preset feature types fails to recover, the recovery judgment module will also correct the current nursing action event to linkage instability. When the duration of continuous co-abnormality reaches the preset duration threshold, it indicates that the co-abnormality of multiple features continues to exist, and the recovery judgment module will also correct the current nursing action event to linkage instability. When none of the three conditions reach the corresponding threshold, the original single event instability judgment result remains unchanged. When any one of the three conditions reaches the threshold, the nursing action event will be output as an enhanced instability judgment result, and the linkage condition type, corresponding value, and feature set involved in the linkage will be written into the result. After obtaining the enhanced instability determination result, the recovery determination module further reads the enhanced instability determination results corresponding to multiple consecutive nursing action events for the same patient, and analyzes the recurrence frequency of similar abnormal features, the transmission amplitude of the residual deviation at the end of the recovery of the previous nursing action event to the baseline time period of the subsequent nursing action event, and the cumulative duration of continuous instability according to the chronological order of the nursing action events on the time axis, so as to output cumulative determination data. In specific implementation, the recovery determination module first sorts the nursing action events for the same patient according to the start time of the action, and pairs the corresponding abnormal features in two adjacent nursing action events. The recurrence frequency is obtained by counting the number of times that similar abnormal features are determined to be delayed recovery, continuous instability, or linked instability in multiple consecutive nursing action events. When the time interval between consecutive events exceeds the maximum correlation interval threshold, the counting starts again. The transmission amplitude is obtained by comparing the residual deviation of the corresponding feature at the end of the recovery of the previous nursing action event with the corresponding feature at the baseline time period of the subsequent nursing action event. The initial deviation is obtained; the continuous instability cumulative duration is obtained by accumulating the duration of delayed recovery, continuous instability, or linked instability of the same abnormal feature in multiple consecutive nursing action events; the unstable interval time between two adjacent nursing action events that is shorter than the interval merging threshold is also included in the continuous instability cumulative duration; after completing the above calculations, the recovery judgment module classifies the local state changes corresponding to each nursing action event according to the cumulative judgment data; when the number of repetitions does not reach the preset number threshold, the transmission amplitude does not reach the preset transmission threshold, and the continuous instability cumulative duration does not reach the preset cumulative threshold, the current nursing action event is judged as a single action instability; when the number of repetitions reaches the preset number threshold, or the transmission amplitude reaches the preset transmission threshold, or the continuous instability cumulative duration reaches the preset cumulative threshold, the current nursing action event is judged as continuous action cumulative instability; finally, the recovery judgment module outputs the final instability judgment result containing sub-item judgment data, linked judgment data, and cumulative judgment data; Through the above implementation process, the recovery judgment module unifies the changes in single features, multi-feature linkage changes, and cumulative changes between multiple consecutive nursing action events within a single nursing action event into a single judgment chain. This expands postoperative nursing monitoring of head and neck surgery from single-point abnormality identification to event-based, linkage-based, and cumulative judgment oriented towards nursing action scenarios, improving the ability to detect rapid deterioration and gradual worsening of local conditions. In practical application: the nursing terminal recorded a turning event at 22:15, a suctioning event at 22:22, and a second turning event at 22:31. The recovery judgment module first calculates the normalized deviation amplitude, recovery duration, and residual deviation for the five types of features corresponding to the turning event at 22:15, and finds that the airflow obstruction feature and the neck pressure transfer feature are both The delayed recovery criteria were met, and a single-event delayed recovery result was generated. Subsequently, it was found that the number of synchronous deviations in this turning event was 3, the number of synchronous delays was 2, and the duration of continuous abnormality reached the preset duration threshold. Therefore, this turning event was corrected as linkage instability. When further analyzing the suctioning event at 22:22 and the turning event again at 22:31, the recovery judgment module found that the airflow obstruction feature recurred in the three nursing action events. Moreover, the residual deviation of airflow obstruction at the end of the recovery of the first turning event remained at a high level in the baseline time period of the second suctioning event and further accumulated into the third turning event. As a result, the number of recurrences, the transmission amplitude, and the cumulative duration of continuous instability all reached the corresponding thresholds. Therefore, the turning event again at 22:31 was judged as continuous action cumulative instability.

[0023] The risk output module associates the instability judgment result with the action type of the corresponding nursing action event, generates the risk level, priority review site and nursing treatment prompts corresponding to the nursing action event, and outputs the postoperative nursing warning result for head and neck surgery. In the postoperative care scenario of head and neck surgery, even if the preceding module has identified nursing action events and completed instability assessment, if the assessment results cannot be further transformed into risk levels, review sites, and treatment prompts that nurses can directly execute, the monitoring results remain at the analysis level and cannot support bedside review and nursing intervention in a timely manner. In particular, under different nursing action triggers, the same instability assessment result is not the same in nursing significance, and the priority review sites and treatment methods corresponding to different abnormal characteristics also differ. Therefore, in this embodiment, the risk output module takes the action type and final instability assessment result corresponding to each nursing action event as input, firstly classifies the risk level based on the action type, instability type, instability duration, and instability accumulation degree; then, it combines the action type and abnormal characteristic type to determine the priority review site and nursing treatment prompts, forming review guidance data; finally, it combines the risk level, priority review site, and nursing treatment prompts to generate a postoperative care warning result for head and neck surgery and outputs it to the nursing terminal. This implementation process includes the following steps: In this embodiment, the risk output module first reads the action type and final instability judgment result corresponding to each nursing action event. The action type includes at least one of the following: turning over, suctioning, dressing change, attempting to swallow, getting up, adjusting pillow position, and traction of drainage tubes. The final instability judgment result includes at least the instability type, instability duration, instability accumulation degree, and the abnormal feature type involved in the judgment. Instability types include single-event delayed recovery, linked instability, and continuous action cumulative instability. The instability duration is the total duration of the abnormal state corresponding to the current nursing action event. The instability accumulation degree is used to characterize the abnormal repetition and accumulation among multiple consecutive nursing action events. The risk output module then establishes a correlation grading rule between action type and instability parameters, and generates a risk level corresponding to each nursing action event based on this rule. In specific implementation, the risk level can be divided into three levels: low risk, medium risk, and high risk. For each nursing action event, the risk level is first determined based on the instability... The system assigns a basic risk score based on the type of event. For example, a single-event delayed recovery corresponds to the first basic score, linked instability corresponds to the second basic score, and continuous action-cumulative instability corresponds to the third basic score. The basic risk score is then adjusted based on the duration of instability. If the duration of instability exceeds the first duration threshold, the first adjustment score is increased; if it exceeds the second duration threshold, the second adjustment score is increased. Further cumulative adjustment is performed based on the degree of instability accumulation. When any of the recurrence frequency, transmission amplitude, or continuous instability accumulation duration reaches the corresponding high threshold, the cumulative adjustment score is further increased. Finally, the risk output module weights the adjusted total risk score according to the action type. For example, the weighting coefficient for actions such as suctioning, attempting swallowing, and turning over can be higher than that for pillow position adjustment. After calculation, the risk output module maps the total risk score to low, medium, or high risk according to a preset grading interval to generate the risk level corresponding to each nursing action event. After obtaining the risk level, the risk output module further determines the priority review sites and nursing intervention prompts corresponding to the current nursing action event based on the action type and the corresponding abnormal feature type in the final instability judgment result, in order to output review guidance data. In specific implementation, the risk output module first establishes a mapping table of "action type - abnormal feature type - review site - intervention prompt". For example, when the action type is turning over or adjusting the pillow position, and the abnormal feature type is mainly neck pressure transfer characteristics and neck temperature difference change characteristics, the priority review site can be determined as the neck pressure area, the area around the skin flap, or the patch coverage area, and the nursing intervention prompt can be determined as checking the head and neck support posture, relieving local pressure, rechecking the pillow position, or re-examining the condition of the skin flap. When the action type is suctioning or attempting to swallow, and the abnormal feature type is mainly airflow obstruction characteristics and local micro-vibration change characteristics, the priority review site can be determined as the mouth and nose area, airway-related areas, or the anterior neck area. Nursing intervention prompts can be identified as verifying ventilation status, checking for secretion retention, assessing whether suctioning is needed again or airway observation needs to be strengthened; when the action type is drainage tube traction or dressing change, and the abnormal feature type is mainly drainage patency and neck compression transfer, the priority verification site can be identified as the area adjacent to the drainage tube, the drainage outlet area, or the area around the wound. Nursing intervention prompts can be identified as checking whether the drainage tube is twisted, compressed, or blocked, and verifying the fixation status and drainage patency; to avoid giving overly narrow verification suggestions based on a single abnormal feature, the risk output module can also merge verification sites and expand intervention prompts according to the combination of abnormal feature types; if the number of abnormal feature types exceeds the preset threshold, they are sorted according to the contribution of each abnormal feature to the final instability judgment result, and only the priority verification sites corresponding to the top two or three items with the highest contribution are retained; finally, the risk output module generates verification guidance data; After receiving the review guidance data, the risk output module further combines the risk level, priority review site, and nursing intervention prompts to generate a post-head and neck surgery nursing early warning result and output it to the nursing terminal. In specific implementation, the risk output module first generates an early warning entry for each nursing action event. Each early warning entry includes at least the patient identifier, nursing action event identifier, action type, action start time, action end time, risk level, priority review site, nursing intervention prompt, and early warning generation time. Subsequently, the output method is determined based on the risk level. When the risk level is low, only the event list on the nursing terminal needs to be updated with a prompt entry; when the risk level is medium, both list prompts and an interface can be displayed simultaneously on the nursing terminal. Pop-up notifications: When the risk level is high, pop-up notifications, color highlighting, and sound alerts can be displayed on the nursing terminal, and the warning item will be marked as a priority. If multiple warning items are generated for the same patient in a short period of time, the risk output module can sort them according to the risk level, the order of warning generation, and the priority of continuous action accumulation instability, and merge adjacent warning items for the same review site. For the output warning items, the nursing terminal can write back the nursing staff's confirmation status, review completion status, and treatment completion status, and the risk output module will associate this written-back information with the corresponding warning item. Finally, the risk output module will continuously output the generated head and neck postoperative nursing warning results to the nursing terminal. Through the above implementation process, the risk output module further transforms the final instability judgment result output by the preceding recovery judgment module into a risk level, priority review site, and nursing intervention prompts that can be executed in nursing care. This allows for the hierarchical expression of abnormal states triggered by different nursing actions, and establishes a correspondence between abnormal characteristics and specific review sites and intervention methods, improving the executability of postoperative head and neck surgery nursing early warning results and the targeted nature of nursing interventions. In practical application: For a postoperative head and neck surgery patient, the recovery judgment module outputs a linked instability result for the "turning over event at 22:15," and determines that the abnormal characteristics are mainly neck compression and transfer features and airflow obstruction features, with an instability duration of 18 seconds, and no... Continuous movement cumulative instability; the risk output module generates a medium-risk level based on the weighted coefficient of the movement corresponding to turning over, the base score corresponding to the linkage instability, and the duration correction score corresponding to the 18-second instability duration. Then, based on the characteristics of neck pressure transfer and airflow obstruction, the neck pressure area and the mouth and nose area are identified as priority review sites, respectively. "Check the head and neck support posture and review the ventilation status" is identified as a nursing intervention prompt, and the above content is combined into a nursing warning result and output to the nursing terminal. If the same patient is judged to have continuous movement cumulative instability in a subsequent suctioning event, the risk output module further upgrades the risk level to high risk and places the warning item in the priority processing position of the nursing terminal.

[0024] Working Principle: This invention revolves around a processing chain that focuses on whether the local condition deteriorates before and after a nursing action. First, the system continuously collects signals such as local temperature, local pressure, micro-vibration, airflow vibration through the mouth and nose, drainage tube patency / discontinuity, and changes in body position using a wearable neck monitoring patch. These signals are then aligned with nursing records entered by caregivers at a terminal, including records of turning over, suctioning, dressing changes, and attempted swallowing, forming raw monitoring data with time and quality markers. Subsequently, using the nursing record time as an anchor point, the system identifies the corresponding nursing action event from the raw monitoring data and further divides each nursing action event into a baseline time period before the action and a disturbance time period during the action. The system analyzes the recovery time after each nursing action and segment. Based on this, it extracts features such as neck temperature difference changes, neck pressure transfer, local micro-vibration changes, airflow obstruction, and drainage patency. It not only determines whether a single feature is abnormal, but also analyzes whether multiple features deteriorate simultaneously, whether recovery is delayed, and whether such abnormalities repeatedly occur and gradually accumulate among multiple consecutive nursing actions. Finally, it outputs the corresponding instability judgment result. Finally, the system combines the result with the nursing action type and converts it into risk level, priority review site, and nursing treatment prompts that nursing staff can directly use, thereby transforming the originally scattered monitoring signals into early warning results for specific nursing actions. For example, during nighttime care of patients after head and neck surgery, nurses first turn the patient over and then perform suctioning. The system continuously collects various monitoring signals before and after the patient's turn, automatically identifies the turning event, and then compares the stable state before turning, the disturbance state during turning, and the recovery state after turning. If the system finds that there is significant neck pressure shift and abnormal increase in airflow and vibration after turning, and these changes do not return to the level before turning in a short time, it will determine that the turning may have caused local instability. If the same abnormality is found to reappear during suctioning, and the abnormal residue from the previous turning has been transmitted to the baseline state before this suctioning, the system will further determine that this is not an occasional fluctuation, but a cumulative risk of deterioration under continuous nursing actions. At this time, the system will not only give an abstract alarm, but directly indicate a high risk level on the nursing terminal and prompt the nurse to prioritize checking the neck pressure area, the oral and nasal area, or the area adjacent to the drainage tube. At the same time, it will provide treatment suggestions such as adjusting the head and neck support posture, checking the ventilation status, and checking whether the drainage is compressed, so that the nurse can locate the problem more quickly and take targeted measures.

[0025] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent wearable head and neck postoperative care monitoring system, characterized in that: include, The data acquisition module collects local temperature signals, local pressure signals, micro-vibration signals, airflow acoustic vibration signals from the neck wearable monitoring patch of patients after head and neck surgery, drainage tube opening and closing signals, and body position change signals. It also collects nursing operation records entered by the nursing terminal and outputs raw monitoring data. The action recognition module aligns the raw monitoring data with the nursing operation records in time, identifies nursing action events corresponding to turning over, suctioning, changing dressings, attempting to swallow, getting up, adjusting pillow position, and pulling drainage tubes, and outputs a sequence of nursing action events including action type, start time, and end time. The window construction module takes each nursing action event in the nursing action event sequence as input, extracts the original monitoring data corresponding to the baseline time period before the action occurs, the disturbance time period during the action, and the recovery time period after the action ends, and outputs a monitoring window that corresponds one-to-one with each nursing action event. The status extraction module extracts features from the raw monitoring data in each monitoring window, calculates the features of neck temperature difference change, neck pressure transfer, local micro-vibration change, airflow obstruction and drainage patency, and outputs the status feature sequence corresponding to each nursing action event. The recovery judgment module compares the disturbance time period and recovery time period in each state feature sequence with the corresponding baseline time period, calculates the state deviation amplitude, recovery duration and residual deviation after recovery, and outputs the instability judgment result corresponding to each nursing action event according to the preset judgment rules. The risk output module associates the instability judgment result with the action type of the corresponding nursing action event, generates the risk level, priority review site and nursing treatment prompts corresponding to the nursing action event, and outputs the postoperative nursing warning result for head and neck surgery.

2. The system of claim 1, wherein: The data acquisition module is used to read the local temperature signal, local pressure signal, micro-vibration signal, airflow acoustic vibration signal from the mouth and nose, drainage tube on / off signal, and body position change signal output by the neck wearable monitoring patch. It adds a uniform timestamp to each signal and rearranges them according to the acquisition time sequence to output multi-channel time-series data. The amplitude continuity, rhythm continuity and channel correlation of each signal within a preset time window are calculated based on multi-channel time series data. Abnormal jump segments, missing segments and cross-channel mismatch segments are marked according to the calculation results to output time series data with quality labels. The data acquisition module is also used to time-align and link the time-series data with quality markers with the nursing operation records entered by the nursing terminal to generate raw monitoring data containing signal values, quality markers and nursing record identifiers.

3. The system of claim 2, wherein: The motion recognition module is used to time-align the body position change signal, local pressure signal, micro-vibration signal, oral and nasal airflow acoustic vibration signal and drainage tube opening and closing signal in the original monitoring data with the nursing operation record. It uses the recording time in the nursing operation record as the anchor point to extract the original monitoring data of the corresponding time period, and generates nursing action candidate segments according to the direction of change, amplitude of change and duration of each signal in the corresponding time period. Based on the combined characteristics of body position change trajectory, local pressure transfer trajectory, micro-vibration rhythm change, oral and nasal airflow acoustic vibration change and drainage tube opening and closing change in each nursing action candidate segment, the nursing action type corresponding to each nursing action candidate segment is determined, and the nursing action identification result is output. The action recognition module is also used to determine the start time and end time of an action based on the start and end boundaries of each nursing action identifier on the time axis, and to associate the action type, start time, and end time to generate a nursing action event sequence.

4. The system of claim 3, wherein: The window construction module is used to read the action type, start time and end time of each nursing action event in the nursing action event sequence, and to determine the baseline time period based on the fluctuation amplitude, signal integrity and nursing record status of the original monitoring data within a continuous preset time before the start time of the action, and to determine the disturbance time period based on the original monitoring data between the start time of the action and the end time of the action, so as to output the baseline time period and the disturbance time period. The window construction module is also used to take the original monitoring data after the end of the action as input, calculate the deviation changes of local temperature signal, local pressure signal, micro-vibration signal, oral and nasal airflow acoustic vibration signal and drainage tube opening and closing signal relative to the baseline time period, and determine the continuous time period when the deviation change is lower than the preset recovery threshold as the recovery time period, so as to output a monitoring window that corresponds one-to-one with each nursing action event.

5. The system of claim 4, wherein: The state extraction module is used to read local temperature signals, local pressure signals, micro-vibration signals, airflow acoustic vibration signals from the mouth and nose, and drainage tube on / off signals within each monitoring window. It then segments the data according to the baseline time period, disturbance time period, and recovery time period, and calculates the mean, rate of change, fluctuation, and difference between adjacent times for the corresponding signals within each time period to output segmented feature data. The state extraction module is also used to calculate the temperature difference change and recovery slope of the local temperature signal between different time periods based on the segmented feature data, so as to generate neck temperature difference change features. Calculate the pressure offset and transfer duration of the local pressure signal between different monitoring locations to generate neck compression transfer characteristics; The amplitude and rhythm changes of the micro-vibration signal are calculated at different time intervals to generate local micro-vibration variation characteristics; Calculate the changes in ventilation amplitude and the proportion of abnormal acoustic vibrations in the airflow acoustic vibration signal from the mouth and nose at different time periods to generate airflow obstruction characteristics; Calculate the number of on / off switching times and duration of continuous blockage of the drainage tube on / off signal in different time periods to generate drainage smoothness characteristics; The state extraction module is also used to correlate the characteristics of neck temperature difference changes, neck pressure transfer, local micro-vibration changes, airflow obstruction, and drainage patency in chronological order to generate a state feature sequence corresponding to each nursing action event.

6. The system of claim 5, wherein: The state extraction module is also used to take each state feature sequence as input, and calculate the start time difference, co-variation duration and recovery order between the neck temperature difference change feature, neck pressure transfer feature, local micro-vibration change feature, airflow obstruction feature and drainage smoothness feature according to the order of appearance of each feature in the same nursing action event, so as to output feature linkage data. The state extraction module is also used to identify the transition moments of each feature from the baseline stable state to the disturbed state and from the disturbed state to the recovery state based on the feature linkage data, and to divide the continuous time period between adjacent transition moments into at least one of the pressure-dominated stage, ventilation obstruction stage, drainage abnormal stage and recovery lag stage, so as to output the state evolution fragments corresponding to the nursing action events. The state extraction module is also used to splice state evolution fragments in chronological order and combine the dominant feature type and duration of each state evolution fragment to generate an enhanced state feature sequence that characterizes the local state change process.

7. The system according to claim 6, characterized in that: The recovery determination module is used to read the neck temperature difference change characteristics, neck pressure transfer characteristics, local micro-vibration change characteristics, airflow obstruction characteristics, and drainage smoothness characteristics corresponding to the baseline time period, disturbance time period, and recovery time period in each state feature sequence. It calculates the deviation of each feature from the baseline time period during the disturbance time period, the recovery duration when it falls back to the baseline fluctuation range during the recovery time period, and the residual deviation from the baseline time period at the end of the recovery period, so as to output the sub-determination data. The recovery judgment module is also used to determine, based on the sub-judgment data, whether the state change corresponding to each feature belongs to one of instantaneous fluctuation, delayed recovery or continuous imbalance, and to generate single-event instability judgment results corresponding to each nursing action event.

8. The system according to claim 7, characterized in that: The recovery determination module is also used to read the sub-determination data corresponding to the same nursing action event, calculate the number of synchronous deviations of each feature during the disturbance period, the number of synchronous lags during the recovery period, and the duration of continuous co-abnormality, so as to output the linkage determination data. The recovery judgment module is also used to correct the single-event instability judgment result based on the linkage judgment data. When the number of synchronization deviations reaches the preset number threshold, the number of synchronization delays reaches the preset number threshold, or the duration of continuous co-abnormality reaches the preset duration threshold, the corresponding nursing action event is corrected to linkage instability to output the enhanced instability judgment result.

9. The system according to claim 8, characterized in that: The recovery determination module is also used to read the enhanced instability determination results corresponding to multiple consecutive nursing action events of the same patient, and calculate the number of recurrences of the same abnormal features, the transmission amplitude of the residual deviation at the end of the recovery of the previous nursing action event to the baseline time period of the next nursing action event, and the cumulative duration of continuous instability according to the order of the nursing action events on the time axis, so as to output the cumulative determination data. The recovery judgment module is also used to determine, based on the accumulated judgment data, whether the local state change corresponding to each nursing action event belongs to either single action instability or continuous action cumulative instability. When the number of repetitions reaches a preset threshold, the transmission amplitude reaches a preset transmission threshold, or the continuous instability cumulative duration reaches a preset cumulative threshold, the current nursing action event is judged as continuous action cumulative instability, and the final instability judgment result is output.

10. The system according to claim 9, characterized in that: The risk output module is used to read the action type and instability judgment result corresponding to each nursing action event, and to perform correlation and classification based on action type, instability type, instability duration and instability accumulation degree to generate the risk level corresponding to each nursing action event; The risk output module is also used to determine the priority review sites and nursing treatment prompts corresponding to the current nursing action event based on the action type and the abnormal feature type corresponding to the instability judgment result, so as to output review guidance data; The risk output module is also used to combine risk level, priority review site and nursing treatment prompts to generate postoperative nursing early warning results for head and neck surgery and output them to the nursing terminal.

Images