Real-time monitoring and early warning method for restlessness during anesthesia recovery period based on video analysis

Abstract

The application discloses a real-time monitoring and early warning method for restlessness during anesthesia recovery based on video analysis, and relates to the technical field of medical monitoring.The technical problem of inaccurate monitoring results and false positives caused by external interference during monitoring of restlessness during anesthesia recovery of a patient is solved.The monitoring video of restlessness during anesthesia recovery is obtained, and continuous monitoring images are obtained by frame picture extraction.The continuous feature sequence is obtained by feature extraction on the continuous monitoring images, the continuous feature sequence is compared with the set non-restlessness baseline and restlessness start threshold to obtain the anesthesia state, the feature of the continuous monitoring images is removed based on the restlessness state to obtain the prediction image, the corresponding restlessness fluctuation parameters are obtained according to the monitoring feature changes in different stages, the restlessness behavior is predicted according to the restlessness feature and the prediction image, the restlessness correlation score is obtained, and the early warning signal is generated based on the restlessness analysis result.

Description

Technical Field

[0001] This invention belongs to the field of medical monitoring, specifically a method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis. Background Technology

[0002] The anesthesia recovery period is a critical stage in which the patient's consciousness and physiological functions gradually recover after general anesthesia. During this stage, patients are very prone to recovery agitation, which manifests clinically as limb movement, restlessness, confusion, and struggling. If these symptoms are not detected and intervened in time, they can easily lead to serious adverse events such as falling out of bed, dislodgement of intravenous lines, extubation of endotracheal tubes, and tearing of surgical incisions, directly threatening the patient's perioperative safety.

[0003] Chinese patent application CN115120219A discloses a method and device for monitoring and alerting agitation. This method involves: acquiring echo signals from a monitoring sensor based on a monitored target; determining whether the monitored target is in motion based on the echo signals; if so, determining whether the duration of the motion meets preset conditions; and if so, generating alarm information corresponding to the monitored target. While this method provides non-contact, real-time monitoring of the monitored target using echo signals from a monitoring sensor, in practical applications, it does not exclude abnormal movements caused by external interference, potentially leading to false alarms or misjudgments in the determination of agitation during the awakening period. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a real-time monitoring and early warning method for agitation during anesthesia recovery based on video analysis, which is used to solve the technical problem in the prior art that the monitoring results are inaccurate and false alarms are caused by external interference when monitoring agitation during the recovery period of anesthetized patients.

[0005] To achieve the above objectives, a first aspect of the present invention provides a method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis, comprising: Step 1: Extract the anesthesia monitoring video and extract the images from the anesthesia monitoring video to obtain continuous monitoring images based on the timestamp sequence; Step 2: Obtain the anesthesia status based on continuous monitoring images, extract agitation features from the continuous monitoring images, and obtain the agitation status based on the anesthesia status and agitation features; Step 3: Based on the agitation state, feature removal is performed on the continuous monitoring images to obtain the predicted images; agitation behavior is predicted based on the agitation features and the predicted images to obtain the agitation analysis results; Step 4: Generate an early warning signal based on the disturbance analysis results.

[0006] Preferably, the step of extracting images from the anesthesia monitoring video to obtain continuous monitoring images based on timestamp order includes: Extract anesthesia monitoring videos; Each frame of the anesthesia monitoring video is bound to a timestamp tag, and the anesthesia monitoring video is then decoded, invalid frames are filtered, and standardized to obtain a standard monitoring video. Frame images are extracted from the standard monitoring video at set time intervals to obtain continuous monitoring images.

[0007] Preferably, the step of acquiring the anesthesia status based on continuous image monitoring includes: Extract continuous monitoring images and their corresponding timestamp labels; Feature extraction is performed on continuous monitoring images to obtain monitoring features. The monitoring features are combined with timestamp labels to obtain a continuous feature sequence of timestamp-detection features. An agitation-free baseline and agitation onset threshold are set based on the clinical agitation score. The clinical agitation score is based on the standard for judging the agitation stage during the recovery period in the database. Based on the time sequence of the timestamp label, the continuous feature sequence is traversed. The feature sequence that breaks through the no-agitation baseline is detected and marked as the detection initial sequence. Taking the detection initial sequence as the starting sequence, the continuous feature sequence is compared with the agitation start threshold to obtain the breakthrough sequence. The anesthesia state corresponding to the breakthrough sequence is marked as the agitation breakthrough state. The anesthesia state of the continuous feature sequence before the starting sequence is marked as the silent state. The anesthesia state of the continuous feature sequence between the starting sequence and the breakthrough sequence is marked as the detection state. The anesthetic state is obtained by integrating the silent state, the detection state, and the agitation breakthrough state.

[0008] Preferably, the step of comparing the continuous feature sequence with the agitation initiation threshold to obtain the breakthrough sequence includes: A1: Extract continuous feature sequences and agitation start thresholds; where the agitation start threshold includes agitation action threshold and agitation time threshold; A2: Determine whether the monitored features in the continuous feature sequence are greater than the agitation threshold; if yes, proceed to A3; if no, extract the monitored features from the next feature sequence and continue the comparison. A3: Record the timestamp label of the current feature sequence and mark it as the start of the disturbance. Return to A2 until the difference between the timestamp label of the continuous feature sequence and the start of the disturbance equals the disturbance time threshold. Mark the feature sequence corresponding to the start of the disturbance as the breakthrough sequence.

[0009] Preferably, the step of obtaining the agitation state based on the anesthesia state and agitation characteristics includes: Extracting the characteristics of anesthesia and agitation; The monitoring features in the continuous feature sequence corresponding to the agitation breakthrough state in the anesthesia state are statistically analyzed and sorted according to the time order of the timestamp labels to obtain the agitation feature fluctuation spectrum. Based on the agitation characteristic fluctuation spectrum, characteristic change curves are obtained. The agitation during the awakening period is divided into agitation stages according to the changes in the monitored characteristics in the characteristic change curves. Based on the changes in the monitored characteristics in different stages of the agitation stage, agitation fluctuation parameters corresponding to different stages are obtained. The agitation state of different stages is obtained based on the agitation fluctuation parameters.

[0010] It should be added that there are corresponding feature thresholds for different stages. If the monitored feature exceeds the feature threshold, it is judged as abnormal agitation. The duration of each stage should not be too long. The closer the monitored feature is to the feature threshold, the shorter the monitoring duration should be. Otherwise, if it is close to the threshold for a long time, abnormal agitation is likely to occur.

[0011] Preferably, the step of obtaining agitation fluctuation parameters corresponding to different stages based on changes in monitoring characteristics at different stages of the agitation phase includes: Extract monitoring features from different stages of the agitation phase; A rectangular two-dimensional coordinate system is established with the timestamp label as the X-axis and the monitoring feature as the Y-axis. In the two-dimensional coordinate system, the monitoring feature corresponding to each timestamp label is marked to obtain the monitoring coordinate point. The coordinate points are connected by a straight line that covers as many monitoring coordinate points as possible. The ratio of the slope to the intercept of this straight line is marked as the turbulence fluctuation parameter. Different stages have different corresponding turbulence fluctuation parameters.

[0012] Preferably, the step of obtaining a predicted image by feature removal from continuously monitored images based on the agitation state includes: Extracting agitated states and continuous monitoring images; The continuous monitoring images are segmented using semantic segmentation and subject extraction techniques to obtain subject images and external region images. The changes in fluctuation parameters during the agitation state are compared with the corresponding external region images. Based on the timestamp labels, it is detected whether there is consistency between the irregular changes in fluctuation parameters and the feature changes in the external region images. External region images with consistency are marked as predicted images.

[0013] Preferably, the step of predicting agitation behavior based on agitation features and predicted images to obtain agitation analysis results includes: Extract the agitation features and their corresponding predicted images; The system detects disturbance-related items in the predicted image, calculates the spatial distance between the disturbance items and the main body in the predicted image using a proportional comparison method, and compares the disturbance time difference between the time of occurrence of the disturbance items and the time of onset of the disturbance action. Based on the disturbance feature fluctuation spectrum, it generates spatiotemporal correlation parameters for different stages, and combines spatial distance, disturbance time difference, and spatiotemporal correlation parameters to generate a disturbance correlation score. Agitation analysis results are generated based on agitation association scores.

[0014] Preferably, the generation of spatiotemporal correlation parameters for different stages based on the agitation characteristic fluctuation spectrum includes: Extracting the wave spectrum of agitation characteristics; The monitoring features corresponding to the agitation feature fluctuation spectrum are sorted in chronological order. The timestamp labels of sudden changes in the monitoring features are detected and marked as abrupt change feature points. The agitation feature fluctuation spectrum is divided into different stages using abrupt change feature points as dividing points to obtain stage feature spectra of different stages. The time length and the difference between the maximum and minimum values ​​in the feature spectrum are statistically analyzed separately. The time weight parameter α and the spatial weight parameter β are defined. The spatiotemporal correlation parameter Sx of the feature spectrum numbered x is calculated by the formula Sx=(αTx+βKx)Kn. Where Tx represents the time length of the feature spectrum numbered x, Kx represents the difference between the maximum and minimum values ​​of the feature spectrum numbered x, Kn represents the normalization parameter, and α+β=1.

[0015] Preferably, generating the early warning signal based on the disturbance analysis results includes: Extract the results of the agitation analysis; A correlation threshold is set, and the agitation correlation score in the agitation analysis results is compared with the correlation threshold. Interference items whose agitation correlation scores exceed the correlation threshold are set as highly interfering items. The monitoring images corresponding to highly interfering items are removed, and the abrupt change feature points in the remaining monitoring images are detected, and warning signals corresponding to the abrupt change feature points are generated. The correlation threshold is set based on the degree of influence of interference items on patients in different stages.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention acquires anesthesia monitoring videos of agitation during the recovery period, extracts frames to obtain continuous monitoring images, extracts features from these images to obtain continuous feature sequences, iterates through these sequences, and compares them with a set agitation-free baseline and agitation onset threshold to determine the anesthesia state. This solves the problems of feature offset and temporal sequence disorder caused by scattered images and asynchronous data, ensuring the data basis for anesthesia state determination is authentic and reliable. Based on the agitation state, features are removed from the continuous monitoring images to obtain predicted images, and corresponding agitation fluctuation parameters are obtained according to the changes in monitoring features at different stages. Agitation behavior is predicted based on agitation features and predicted images, and agitation correlation scores are obtained. This can accurately remove interference, purify the true agitation signal, and quantify parameters in stages, conforming to the agitation development pattern. Based on the agitation analysis results, an early warning signal is generated, eliminating feature changes caused by interference and improving monitoring robustness.

[0017] 2. This invention segments and divides monitoring images, and combines temporal and spatial changes of monitoring features in the images with spatial location information to simultaneously analyze feature fluctuations and the spatiotemporal distribution of interference. It takes into account both the temporal continuity of agitated features and the degree of influence of spatial interference, making it suitable for complex clinical scenarios. Even in environments with multiple interferences, it can still output reliable analysis results stably. The monitoring images after interference elimination are used to detect abnormal feature changes and generate corresponding early warning signals, making the early warning more targeted, reducing invalid and missed early warnings, and improving the practicality of the entire monitoring and early warning system. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating the working principle of the present invention.

[0020] Figure 2 This is a flowchart illustrating a process in one embodiment of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.

[0022] Please see Figure 1 The first aspect of the present invention provides a method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis, comprising: Step 1: Extract the anesthesia monitoring video and extract the images from the anesthesia monitoring video to obtain continuous monitoring images based on the timestamp sequence.

[0023] For example, an anesthesia monitoring video is captured in real time by a camera placed above the operating room and stored in a storage system. The camera's recording range must cover the operating table and the nearby movable area. The recording start time can be 10 minutes after anesthesia and surgery or other custom time. The recording start time must be within the time when the patient is still under anesthesia and has not started to agitate.

[0024] In this embodiment, recording begins at 3 PM and lasts for a total of 1 hour. The anesthesia monitoring video is extracted at 10-frame intervals, and the extracted frames are named using the extraction time as a timestamp. The quality of the extracted frames is verified, and invalid or blurry images are removed through video decoding, invalid frame filtering, and standardization. Finally, a continuous monitoring image based on the timestamp sequence is formed.

[0025] It should be added that when a camera is recording, it needs to determine the subject being recorded, which can be done through identity authentication or by using the center of the camera as the subject.

[0026] Step 2: Obtain the anesthesia status based on continuous monitoring images, extract agitation features from the continuous monitoring images, and obtain the agitation status based on the anesthesia status and agitation features.

[0027] For example, monitoring features are obtained by extracting features from continuous monitoring images through inter-frame motion feature calculation. The monitoring features specifically include the positions of different limbs of the patient in the frame. The timestamp labels corresponding to the continuous monitoring images are bound to the monitoring features to obtain a continuous feature sequence of timestamp-detection features.

[0028] In this embodiment, the baseline for agitation and the threshold for the onset of agitation are set based on the clinical agitation score. The clinical agitation score adopts the industry-standard Riker Sedation-Agitation Scale (SAS) as the judgment criterion. Specifically, the SAS score of 1-2 points indicates agitation (quiet, drowsy); the SAS score of 3 points indicates a borderline state (able to cooperate with commands, with occasional slight limb movements); and the SAS score of 4-7 points indicates agitation (4 points: restlessness, attempting to sit up; 5-7 points: violent struggle, pulling out of tubes). Therefore, the baseline for agitation in this embodiment is 2 points, and the threshold for the onset of agitation is 4 points.

[0029] Furthermore, due to differences in patients' age, gender, and anesthetic dosage, the established agitation baseline and agitation onset threshold can be adjusted according to the actual situation.

[0030] The continuous feature sequences are traversed in ascending order of timestamps, and segmented and marked according to the no-anxiety baseline and the agitation start threshold. Starting from 15:10:00.000, the subsequent sequences are traversed. During the period from 15:00:00 to 15:20:00, the monitored features are below the no-anxiety baseline and are marked as silent. From 15:20:00 to 15:25:00, the monitored features exceed the no-anxiety baseline but do not reach the agitation start threshold and are marked as detected. After 15:25:00, the monitored features exceed the agitation start threshold and are marked as agitation breakthrough state.

[0031] The feature comparison process specifically involves determining whether the monitored features in the continuous feature sequence are greater than the agitation threshold. If so, the timestamp label of the current feature sequence is recorded and marked as the agitation start label. The process continues to determine whether the monitored features in the continuous feature sequence of the next timestamp label are greater than the agitation threshold until the difference between the timestamp label of the continuous feature sequence and the agitation start label equals the agitation time threshold. The feature sequence corresponding to the agitation start label is then marked as the breakthrough sequence. If not, the monitored features in the next feature sequence are extracted and the comparison continues.

[0032] It should be added that the timestamp label corresponding to the breakthrough sequence indicates the time when the patient's agitation began during the awakening period. The agitation that the patient exhibited during the detection state was relatively small and was in the period between awakening and agitation. When the monitored features reached the agitation initiation threshold, it indicated that agitation had begun.

[0033] The monitored features in the continuous feature sequence after 15:25:00 are extracted and statistically analyzed, and sorted according to the time order of the timestamp labels to obtain the agitation feature fluctuation spectrum. Based on the agitation feature fluctuation spectrum, a corresponding feature change curve graph based on a two-dimensional rectangular coordinate system is generated. The slope change, peak nodes, and fluctuation amplitude in the feature change curve graph are detected. The feature change curve graph is divided according to the slope change, peak nodes, and fluctuation amplitude, and the awakening agitation is divided into several stages based on the timestamp labels corresponding to the changes. In this embodiment, the awakening agitation is divided into four stages according to the slope change, peak nodes, and fluctuation amplitude: the agitation precursor stage between 15:25:00 and 15:30:00, the agitation rising stage between 15:30:00 and 15:35:00, the agitation peak stage between 15:35:00 and 15:40:00, and the agitation falling stage between 15:40:00 and 15:45:00.

[0034] It should be noted that when detecting changes in slope, peak nodes, and fluctuation amplitude in the characteristic change curve, if abnormal data such as irregular or unexpected slope changes occur, it indicates that the patient's actions are not in a stable and continuous awakening agitation process, but rather a sudden, irregular, and atypical limb movements or physiological reactions, requiring the immediate generation of an emergency warning signal.

[0035] A rectangular two-dimensional coordinate system is established with the timestamp label as the X-axis and the monitoring feature as the Y-axis. In the two-dimensional coordinate system, the monitoring feature corresponding to each timestamp label is marked to obtain the monitoring coordinate point. The coordinate points are connected by a straight line that covers as many monitoring coordinate points as possible. The ratio of the slope to the intercept of this straight line is marked as the agitation fluctuation parameter. In this embodiment, the agitation fluctuation parameter is 0.1 in the agitation precursor stage, 0.6 in the agitation rise stage, 0.25 in the agitation peak stage, and 0.35 in the agitation fall stage. The agitation fluctuation parameter is then integrated with the corresponding agitation stage to obtain the agitation state.

[0036] It should be added that there are corresponding feature thresholds for different stages. If the monitored feature exceeds the feature threshold, it is judged as abnormal agitation. The duration of each stage should not be too long. The closer the monitored feature is to the feature threshold, the shorter the monitoring duration should be. Otherwise, if it is close to the threshold for a long time, abnormal agitation is likely to occur.

[0037] Step 3: Based on the agitation state, feature removal is performed on the continuous monitoring images to obtain the predicted images; agitation behavior is predicted based on the agitation features and the predicted images to obtain the agitation analysis results.

[0038] For example, semantic segmentation and subject extraction techniques are used to segment continuously monitored images, resulting in subject images and external region images. The subject image is the patient image, and the external region images are images of other regions in the continuously monitored images besides the patient. The changes in agitation fluctuation parameters corresponding to each stage of the agitation state are temporally aligned and compared with the feature changes of the external region images at the same time stamp. The feature changes of the external region images include changes in area, positional movement, sudden grayscale changes, contour jitter, and changes in occlusion range. The changes in agitation fluctuation parameters include irregular changes such as sudden slope changes, abnormal peak values, irregular fluctuation amplitude, and curve jumps. Based on the time stamp label, the irregular changes in agitation fluctuation parameters and the feature changes of the external region images are detected segment by segment to see if they occur simultaneously in the same time period, whether the change trends are consistent, and whether the durations overlap. If the two occur synchronously in time, have consistent change trends, and highly overlap in duration, they are determined to be consistent. In this case, the part of the external region image with feature changes corresponding to the time stamp label is identified as the source of interference causing agitation feature abnormalities, and this part of the external region image is marked as the predicted image.

[0039] To detect disturbances related to agitation in the predicted image, in this embodiment, a change in agitation fluctuation parameters was detected at 15:38:00. The detected disturbance was the movement of medical staff. The image was then normalized to its spatial scale using a proportional comparison method. The actual physical distance was calculated based on the pixel ratio. The spatial distance between each disturbance in the predicted image and the patient was calculated to be 2 meters. The times of appearance, duration, and disappearance of the disturbance were recorded and compared with the start time of the agitation. The disturbance time difference was found to be 10 seconds. In this embodiment, the time weight parameter α = 0.6 and the spatial weight parameter β = 0.4 were defined. The spatiotemporal correlation parameter Sx = 0.9 for the feature spectrum of stage x was calculated using the formula Sx = (αTx + βKx)Kn, where Kn = 0.1.

[0040] The spatial distance, interference time difference, and spatiotemporal correlation parameters are multiplied by their corresponding weight parameters, summed, and normalized to ten to obtain the agitation correlation score. The higher the agitation correlation score, the greater the correlation between the interference and the patient's actions. The interference items and the corresponding agitation correlation scores are integrated to obtain the agitation analysis results.

[0041] Step 4: Generate an early warning signal based on the disturbance analysis results.

[0042] For example, in this embodiment, the correlation threshold for the peak stage of agitation is set to 0.8. The agitation correlation score corresponding to each interference item in the agitation analysis results is compared with the correlation threshold of the corresponding agitation stage one by one: if the agitation correlation score is greater than the correlation threshold of the corresponding stage, the interference item is marked as a highly interference item. It is determined that such interference items will seriously confuse the real agitation characteristics, and all monitoring image frames corresponding to the highly interference items are directly removed, while the pure monitoring image sequence without highly interference is retained; if the agitation correlation score is less than or equal to the correlation threshold of the corresponding stage, it is determined as an ordinary interference item, and the corresponding monitoring image is retained with only a mark and note.

[0043] Traverse the agitation feature fluctuation spectrum corresponding to the time sequence, detect drastic feature points in the remaining image sequence, locate the real agitation outbreak node, and generate the corresponding early warning signal.

[0044] It should be added that, when generating early warning signals, the corresponding early warning signals are generated in different levels according to the agitation stage, characteristic fluctuation amplitude, and duration of the agitation characteristic points. The early warning level is strongly linked to the agitation risk and clinical intervention needs, which can generate more refined early warning signals.

[0045] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0046] Working principle of the invention: This invention acquires anesthesia monitoring videos of agitation during the recovery period, extracts frames to obtain continuous monitoring images, extracts features from these images to obtain continuous feature sequences, iterates through these sequences, and compares them with a set agitation-free baseline and agitation onset threshold to obtain the anesthesia state. Based on the agitation state, features are removed from the continuous monitoring images to obtain predicted images, and corresponding agitation fluctuation parameters are obtained according to changes in monitoring features at different stages. Agitation behavior is predicted based on agitation features and predicted images, and an agitation association score is obtained. By segmenting the monitoring images and combining the spatiotemporal changes of monitoring features in the images with temporal feature changes and spatial location information, the spatiotemporal distribution of feature fluctuations and interference is analyzed simultaneously. The monitoring images after interference removal are used to detect abnormal feature changes and generate corresponding early warning signals.

[0047] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis, characterized in that, include: Step 1: Extract the anesthesia monitoring video and extract the images from the anesthesia monitoring video to obtain continuous monitoring images based on the timestamp sequence; Step 2: Obtain the anesthesia status based on continuous monitoring images, extract agitation features from the continuous monitoring images, and obtain the agitation status based on the anesthesia status and agitation features; Step 3: Based on the agitation state, feature removal is performed on the continuous monitoring images to obtain the predicted images; agitation behavior is predicted based on the agitation features and the predicted images to obtain the agitation analysis results; Step 4: Generate an early warning signal based on the disturbance analysis results; The method of obtaining a predicted image by feature removal from continuously monitored images based on agitation status includes: Extracting agitated states and continuous monitoring images; The continuous monitoring images are segmented using semantic segmentation and subject extraction techniques to obtain subject images and external region images. The changes in fluctuation parameters in the agitated state are compared with the corresponding external region images. Based on the timestamp label, it is detected whether there is consistency between the irregular changes in fluctuation parameters and the feature changes in the external region images. External region images with consistency are marked as predicted images. The agitation analysis results obtained by predicting agitation behavior based on agitation features and predicted images include: Extract the agitation features and their corresponding predicted images; The system detects disturbance-related items in the predicted image, calculates the spatial distance between the disturbance items and the main body in the predicted image using a proportional comparison method, and compares the disturbance time difference between the time of occurrence of the disturbance items and the time of onset of the disturbance action. Based on the disturbance feature fluctuation spectrum, it generates spatiotemporal correlation parameters for different stages, and combines spatial distance, disturbance time difference, and spatiotemporal correlation parameters to generate a disturbance correlation score. Agitation analysis results are generated based on agitation association scores; The generation of spatiotemporal correlation parameters for different stages based on the agitation characteristic fluctuation spectrum includes: Extracting the wave spectrum of agitation characteristics; The monitoring features corresponding to the agitation feature fluctuation spectrum are sorted in chronological order. The timestamp labels of sudden changes in the monitoring features are detected and marked as abrupt change feature points. The agitation feature fluctuation spectrum is divided into different stages using abrupt change feature points as dividing points to obtain stage feature spectra of different stages. The time length and the difference between the maximum and minimum values ​​in the feature spectrum are statistically analyzed separately. The time weight parameter α and the spatial weight parameter β are defined. The spatiotemporal correlation parameter Sx of the feature spectrum numbered x is calculated by the formula Sx=(αTx+βKx)Kn. Where Tx represents the time length of the feature spectrum numbered x, Kx represents the difference between the maximum and minimum values ​​of the feature spectrum numbered x, Kn represents the normalization parameter, and α+β=1.

2. The method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis according to claim 1, characterized in that, The step of extracting images from the anesthesia monitoring video to obtain continuous monitoring images based on timestamp order includes: Extract anesthesia monitoring videos; Each frame of the anesthesia monitoring video is bound to a timestamp tag, and the anesthesia monitoring video is then decoded, invalid frames are filtered, and standardized to obtain a standard monitoring video. Frame images are extracted from the standard monitoring video at set time intervals to obtain continuous monitoring images.

3. The method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis according to claim 1, characterized in that, The method of acquiring anesthesia status based on continuous image monitoring includes: Extract continuous monitoring images and their corresponding timestamp labels; Feature extraction is performed on continuous monitoring images to obtain monitoring features. The monitoring features are combined with timestamp labels to obtain a continuous feature sequence of timestamp-detection features. An agitation-free baseline and agitation onset threshold are set based on the clinical agitation score. The clinical agitation score is based on the standard for judging the agitation stage during the recovery period in the database. Based on the time sequence of the timestamp label, the continuous feature sequence is traversed. The feature sequence that breaks through the no-agitation baseline is detected and marked as the detection initial sequence. Taking the detection initial sequence as the starting sequence, the continuous feature sequence is compared with the agitation start threshold to obtain the breakthrough sequence. The anesthesia state corresponding to the breakthrough sequence is marked as the agitation breakthrough state. The anesthesia state of the continuous feature sequence before the starting sequence is marked as the silent state. The anesthesia state of the continuous feature sequence between the starting sequence and the breakthrough sequence is marked as the detection state. The anesthetic state is obtained by integrating the silent state, the detection state, and the agitation breakthrough state.

4. The method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis according to claim 3, characterized in that, The step of comparing the continuous feature sequence with the agitation initiation threshold to obtain the breakthrough sequence includes: A1: Extract continuous feature sequences and agitation start thresholds; where the agitation start threshold includes agitation action threshold and agitation time threshold; A2: Determine whether the monitored features in the continuous feature sequence are greater than the agitation threshold; if yes, proceed to A3; if no, extract the monitored features from the next feature sequence and continue the comparison. A3: Record the timestamp label of the current feature sequence and mark it as the start of the disturbance. Return to A2 until the difference between the timestamp label of the continuous feature sequence and the start of the disturbance equals the disturbance time threshold. Mark the feature sequence corresponding to the start of the disturbance as the breakthrough sequence.

5. The method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis according to claim 1, characterized in that, The process of obtaining the agitation state based on the anesthesia state and agitation characteristics includes: Extracting the characteristics of anesthesia and agitation; The monitoring features in the continuous feature sequence corresponding to the agitation breakthrough state in the anesthesia state are statistically analyzed and sorted according to the time order of the timestamp labels to obtain the agitation feature fluctuation spectrum. Based on the agitation characteristic fluctuation spectrum, characteristic change curves are obtained. The agitation during the awakening period is divided into agitation stages according to the changes in the monitored characteristics in the characteristic change curves. Based on the changes in the monitored characteristics in different stages of the agitation stage, agitation fluctuation parameters corresponding to different stages are obtained. The agitation state of different stages is obtained based on the agitation fluctuation parameters.

6. The method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis according to claim 5, characterized in that, The method of obtaining agitation fluctuation parameters corresponding to different stages based on changes in monitoring characteristics at different stages of the agitation phase includes: Extract monitoring features from different stages of the agitation phase; A rectangular two-dimensional coordinate system is established with the timestamp label as the X-axis and the monitoring feature as the Y-axis. In the two-dimensional coordinate system, the monitoring feature corresponding to each timestamp label is marked to obtain the monitoring coordinate point. The coordinate points are connected by a straight line that covers as many monitoring coordinate points as possible. The ratio of the slope to the intercept of this straight line is marked as the turbulence fluctuation parameter. Different stages have different corresponding turbulence fluctuation parameters.

7. The method for real-time monitoring and early warning of agitation during anesthesia recovery based on video analysis according to claim 1, characterized in that, The generation of early warning signals based on the disturbance analysis results includes: Extract the results of the agitation analysis; A correlation threshold is set, and the agitation correlation score in the agitation analysis results is compared with the correlation threshold. Interference items whose agitation correlation scores exceed the correlation threshold are set as highly interfering items. The monitoring images corresponding to highly interfering items are removed, and the abrupt change feature points in the remaining monitoring images are detected, and warning signals corresponding to the abrupt change feature points are generated. The correlation threshold is set based on the degree of influence of interference items on patients in different stages.

Images