Old patient postoperative delirium early warning method and system based on behavior pattern recognition

Abstract

The invention discloses an old patient postoperative delirium early warning method and system based on behavior pattern recognition. According to the method, a patient video is collected through ward monitoring equipment, a target area image is extracted through deep learning, behavior restlessness is quantified in combination with a Harris angular point and an optical flow method, facial expressions, fine limb jitter and other detail action changes are captured by analyzing matching failure key points, and the real-time delirium expression degree is obtained through fusion of the two. And fitting the performance degrees of all moments into a time sequence curve, extracting delirium attack curve segments exceeding a threshold value, calculating the product of skewness and an accumulated value of each curve segment as severity, analyzing based on severity trend and attack duration change to obtain an aggravation factor, and finally comprehensively generating the comprehensive severity of postoperative delirium. The system carries out automatic risk grading and early warning according to a preset threshold value, differentiated intervention guidance is provided for medical staff, early, dynamic and accurate early warning of postoperative delirium is achieved, and the early warning accuracy and clinical practicability are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology, and more specifically, to a method and system for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. Background Technology

[0002] Developing a behavioral pattern recognition-based early warning system for postoperative delirium in elderly patients is a complex and multidisciplinary problem, involving clinical medicine, artificial intelligence, behavioral science, and nursing. Postoperative delirium (POD) is an acute, fluctuating cognitive impairment syndrome characterized by significant changes in consciousness, attention, thinking, and perception. Clinical manifestations include confusion, inattention, disorganization, mood swings, and hallucinations. POD is a common postoperative complication in elderly patients, with a high incidence and poor prognosis, placing a significant burden on both patients and the healthcare system.

[0003] Because elderly patients have significant individual differences and their behavioral patterns may vary from person to person, it is difficult to establish a unified behavioral pattern model. The lack of unified standardized assessment criteria limits the applicability of the model to different individuals. Furthermore, the behavioral patterns of elderly patients may become complex and unstable due to various factors (such as cognitive decline, changes in physical condition, etc.), which may lead to a large error in the postoperative delirium warning of elderly patients based on behavioral pattern recognition.

[0004] Existing technologies have the following main drawbacks: Due to the large individual differences and complex and variable behavioral patterns of elderly patients, existing methods are difficult to establish universal models, resulting in poor generalization ability and insufficient standardization; they usually focus on macroscopic movements and cannot effectively capture early key signs such as facial expressions and fine tremors, thus having limited sensitivity; at the same time, existing technologies lack in-depth analysis of the fluctuation and temporal evolution of delirium symptoms, making it difficult to distinguish between transient behavioral abnormalities and progressive deterioration trends; in addition, many methods have weak interpretability, and the correlation between early warning results and specific clinical interventions is unclear, reducing their practical value and credibility in clinical decision-making. Summary of the Invention

[0005] The main objective of this invention is to provide a method and system for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition, thereby addressing at least the problems of poor model generalization ability and weak ability to recognize subtle features in existing technologies. This significantly improves the accuracy and sensitivity of early warning for postoperative delirium in elderly patients.

[0006] To achieve the above objectives, a method and system for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition are provided.

[0007] In a first aspect, the present invention provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition, the method comprising:

[0008] Collect videos of the target patient's daily activities and obtain images of the target region in each frame of the videos;

[0009] The behavioral agitation of the target patient is quantified based on multiple target region images, and the amount of detailed changes in the target patient's movements is calculated. The degree of postoperative delirium in the target patient is calculated based on the behavioral agitation and the amount of detailed changes in movements.

[0010] The severity of postoperative delirium was fitted to a postoperative delirium curve, and multiple curve segments in which postoperative delirium occurred were extracted from the postoperative delirium curve. Multiple postoperative delirium severity was calculated based on multiple curve segments in which postoperative delirium occurred. Postoperative delirium aggravation factor was calculated based on multiple curve segments in which postoperative delirium occurred and multiple postoperative delirium severity. The overall severity of postoperative delirium was calculated based on multiple postoperative delirium severity and postoperative delirium aggravation factor.

[0011] A severity threshold is preset, and the severity warning is obtained by comparing the overall severity of postoperative delirium with the severity threshold.

[0012] Specifically, videos of the target patient's daily activities are collected, and images of the target region are obtained from each frame of the daily activity videos, including:

[0013] The surveillance equipment in the ward where the target patient is located was used to collect videos of the target patient's daily activities without any blind spots;

[0014] Remove invalid video segments from daily activity videos that do not contain the target patient, and retain valid video segments that contain the target patient;

[0015] A human target segmentation algorithm based on a deep convolutional neural network is used to obtain the target region image in each frame of a valid video segment, where each frame of the video image corresponds to a timestamp.

[0016] Specifically, behavioral agitation of the target patient is quantified based on images of multiple target regions, including:

[0017] The Harris corner detection method is used to extract multiple key points of the target region image in each frame of video image, and the Harris response value corresponding to multiple key points is calculated. The key points include the head, torso and limbs of the target patient.

[0018] Optical flow is used to match key points of a target region in consecutive frames of video images to obtain multiple matching pairs and multiple motion vectors for the matching pairs.

[0019] The motion speed of the target patient in adjacent video frames is calculated based on multiple motion vectors, multiple Harris response values, and the consistency of the motion directions corresponding to the multiple motion vectors.

[0020] The number of connected components is calculated from multiple target region images, and the range of motion of the target patient is calculated from the positional changes of the target regions and the number of connected components.

[0021] Behavioral agitation is calculated based on the speed and amplitude of the movement.

[0022] Specifically, the detailed changes in motion of the target patient are calculated based on images of multiple target regions, including:

[0023] Identify multiple key points of failed matching, calculate the number and importance of multiple key points of failed matching, and calculate the amount of change in detailed actions based on the number and importance of multiple key points of failed matching.

[0024] Among them, several key points of failed matching reflected the target patient's facial expressions and subtle limb tremors.

[0025] Specifically, the severity of postoperative delirium in target patients was calculated based on the degree of behavioral agitation and changes in detail of motor function, including:

[0026] The mean values ​​of behavioral agitation and corresponding changes in detailed movements at each moment were calculated to obtain the degree of postoperative delirium.

[0027] Specifically, the degree of postoperative delirium was fitted to a postoperative delirium performance curve, and multiple curve segments showing the occurrence of postoperative delirium were extracted from the postoperative delirium performance curve, including:

[0028] The degree of delirium at all times was fitted into a postoperative delirium curve in chronological order.

[0029] A preset threshold for the degree of delirium was set. Multiple postoperative delirium levels were compared with the threshold. All postoperative delirium levels exceeding the threshold were extracted from the postoperative delirium curve as the curve segments in which postoperative delirium occurred.

[0030] Specifically, the severity of multiple postoperative delirium symptoms was calculated based on the curve segments showing multiple postoperative delirium symptoms, including:

[0031] Calculate the skewness of the curve segments showing multiple postoperative delirium phenomena;

[0032] Calculate the cumulative value of the severity of multiple postoperative delirium symptoms along the curve segments;

[0033] The severity of postoperative delirium is obtained by multiplying multiple skewnesses by the cumulative values ​​of the manifestations corresponding to the multiple skewnesses.

[0034] Specifically, the aggravation factor of postoperative delirium was calculated based on multiple curve segments of postoperative delirium and multiple severity levels of postoperative delirium, including:

[0035] The first slope was obtained by linear fitting of the severity of multiple postoperative delirium manifestations;

[0036] Clustering the time intervals of multiple postoperative delirium occurrence curve segments to obtain the duration of multiple delirium episodes, and fitting multiple durations to obtain a second slope.

[0037] The aggravation factor of postoperative delirium was calculated based on the first slope and the second slope.

[0038] Specifically, the overall severity of postoperative delirium was calculated based on multiple postoperative delirium severity and aggravating factors, including:

[0039] The total severity of postoperative delirium is obtained by summing the severity of multiple postoperative delirium symptoms.

[0040] The overall severity of postoperative delirium was obtained by multiplying the total severity of postoperative delirium by the aggravation factor and normalizing the result.

[0041] Secondly, the present invention provides a postoperative delirium early warning system for elderly patients based on behavioral pattern recognition. The system applies the method described in the first part and includes:

[0042] The data acquisition and preprocessing unit is used to acquire videos of the target patient's daily activities and obtain images of the target region in each frame of the daily activity videos.

[0043] The feature extraction and postoperative delirium severity calculation unit is connected to the data acquisition and preprocessing unit. The feature extraction and postoperative delirium severity calculation unit is used to quantify the behavioral agitation of the target patient based on multiple target region images and to calculate the amount of change in the detailed movements of the target patient. The postoperative delirium severity of the target patient is calculated based on the behavioral agitation and the amount of change in the detailed movements.

[0044] The temporal analysis and comprehensive severity assessment unit is connected to the feature extraction and postoperative delirium severity calculation unit. The temporal analysis and comprehensive severity assessment unit is used to fit the postoperative delirium severity into a postoperative delirium performance curve, extract multiple curve segments of postoperative delirium occurrence from the postoperative delirium performance curve, calculate multiple postoperative delirium severity based on multiple postoperative delirium occurrence curve segments, calculate postoperative delirium aggravation factor based on multiple postoperative delirium occurrence curve segments and multiple postoperative delirium severity, and calculate the comprehensive severity of postoperative delirium based on multiple postoperative delirium severity and postoperative delirium aggravation factor.

[0045] The risk grading and early warning response unit is connected to the time series analysis and comprehensive severity assessment unit. The risk grading and early warning response unit is used to preset severity thresholds and compare the comprehensive severity of postoperative delirium with the severity threshold to obtain the corresponding severity warning.

[0046] This application provides a method and system for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method first collects videos of the patient's daily activities from ward monitoring equipment without blind spots, and then uses deep learning technology to extract images of the patient's body regions from the videos. Subsequently, by analyzing the speed and amplitude of the patient's movements, as well as subtle changes in facial expressions and limb tremors, the method quantifies the patient's agitation and changes in subtle movements, thereby calculating the degree of postoperative delirium at each moment. Next, based on time-series analysis, a delirium performance curve is fitted to extract the delirium onset time period, and its severity and worsening trend are comprehensively assessed, ultimately obtaining a comprehensive severity level reflecting the current delirium risk. The system automatically performs risk classification according to preset thresholds and outputs corresponding warnings, providing medical staff with differentiated decision support from intensive observation to emergency intervention, achieving early, dynamic, and accurate early warning of postoperative delirium. Attached Figure Description

[0047] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 A flowchart illustrating a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition, as provided in this application;

[0049] Figure 2 A schematic diagram illustrating the method for removing invalid video segments from which no target patient exists in a behavioral pattern recognition-based early warning method for postoperative delirium in elderly patients provided in this application;

[0050] Figure 3 A schematic diagram of the skewness of a postoperative delirium early warning method for elderly patients based on behavioral pattern recognition provided in this application;

[0051] Figure 4 This is a connection diagram of a postoperative delirium early warning system for elderly patients based on behavioral pattern recognition, provided in this application. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein.

[0054] In this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0055] This application provides a method and system for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method continuously collects videos of patients' daily activities using monitoring equipment deployed in the ward, and extracts images of target areas of the patient using deep learning technology for behavioral analysis. The system first quantifies the overall agitation level and subtle changes in the patient's movements, comprehensively calculating the real-time severity of delirium. Subsequently, it analyzes the fluctuation trend and episodic characteristics of delirium through temporal modeling, assessing its severity and tendency to worsen, and finally generating a comprehensive severity index reflecting the current risk of delirium. Based on a preset risk threshold, the system automatically outputs graded early warnings, assisting medical staff in making precise nursing decisions from daily observation to emergency intervention, achieving early detection and dynamic early warning of postoperative delirium.

[0056] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0057] Figure 1 A flowchart illustrating a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition, as provided in this application, is shown below. Figure 1 As shown in this embodiment, a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition is provided. The method includes:

[0058] Collect videos of the target patient's daily activities and obtain images of the target region in each frame of the videos;

[0059] The behavioral agitation of the target patient is quantified based on multiple target region images, and the amount of detailed changes in the target patient's movements is calculated. The degree of postoperative delirium in the target patient is calculated based on the behavioral agitation and the amount of detailed changes in movements.

[0060] The severity of postoperative delirium was fitted to a postoperative delirium curve, and multiple curve segments in which postoperative delirium occurred were extracted from the postoperative delirium curve. Multiple postoperative delirium severity was calculated based on multiple curve segments in which postoperative delirium occurred. Postoperative delirium aggravation factor was calculated based on multiple curve segments in which postoperative delirium occurred and multiple postoperative delirium severity. The overall severity of postoperative delirium was calculated based on multiple postoperative delirium severity and postoperative delirium aggravation factor.

[0061] A severity threshold is preset, and the severity warning is obtained by comparing the overall severity of postoperative delirium with the severity threshold.

[0062] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method involves collecting videos of elderly patients' daily activities after surgery, extracting target regions from each frame, and quantifying the patient's agitation and changes in subtle movements to calculate the severity of postoperative delirium. The severity is then fitted to a curve, identifying multiple segments of the curve that indicate delirium, and calculating the severity of each segment and the overall aggravation factor. Finally, a comprehensive severity assessment of postoperative delirium is obtained. By comparing this comprehensive value with a preset severity threshold, automated early warning of postoperative delirium is achieved.

[0063] This method enables objective, continuous, and non-invasive monitoring of postoperative delirium. By combining behavioral agitation with changes in subtle motor movements, it can more comprehensively and sensitively identify early manifestations of delirium. The curve segment-based analysis helps to capture the fluctuation characteristics of delirium, while the calculation of comprehensive severity and aggravation factors improves the accuracy and timeliness of early warning, providing a reliable basis for clinical intervention.

[0064] Specifically, videos of the target patient's daily activities are collected, and images of the target region are obtained from each frame of the daily activity videos, including:

[0065] The surveillance equipment in the ward where the target patient is located was used to collect videos of the target patient's daily activities without any blind spots;

[0066] Remove invalid video segments from daily activity videos that do not contain the target patient, and retain valid video segments that contain the target patient;

[0067] A human target segmentation algorithm based on a deep convolutional neural network is used to obtain the target region image in each frame of a valid video segment, where each frame of the video image corresponds to a timestamp.

[0068] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method collects videos of patients' daily activities through a full-coverage monitoring device deployed in the ward, and automatically removes invalid segments that do not contain the patient, retaining only the valid video segments in which the target patient appears. Subsequently, a human target segmentation algorithm based on a deep convolutional neural network is used to process each frame of the valid video segment, accurately extracting the target region image with the patient as the main body, and marking the corresponding timestamp for each frame of the image, thereby realizing the structured and targeted processing of video data.

[0069] In this solution, the monitoring system is deployed in the ward and uses cameras to capture patients' daily activities. By analyzing the patients' monitoring video data in real time, the system analyzes the changes in patients' movements, postures and behaviors, and identifies abnormal behavioral patterns related to delirium.

[0070] Therefore, an arbitrary elderly patient was selected within the hospital as the target elderly patient. Video data was collected using monitoring equipment in the ward where the target elderly patient was located, at a rate of one frame per second. Then, a human target segmentation algorithm based on a Deep Convolutional Neural Network (DCNN) was used to obtain the target region of the target elderly patient in each frame of the video image. Each video frame corresponds to a timestamp.

[0071] In this process, invalid video segments that do not contain the target elderly patient (such as when the patient leaves the ward or goes to the restroom) are removed, while valid video segments that do contain the target elderly patient are retained.

[0072] Figure 2 A schematic diagram illustrating the method for removing invalid video segments from which no target patient exists in a behavioral pattern recognition-based early warning method for postoperative delirium in elderly patients provided in this application, as shown in the figure. Figure 2 As shown, invalid video segments that do not contain the target elderly patients are directly extracted and deleted to eliminate invalid video segments that do not contain the target elderly patients.

[0073] A pre-trained Deep Convolutional Neural Network (DCNN) human segmentation model is employed. The model's input is each frame of the original video image from a valid video segment. Each frame is fed into the DCNN model. The network automatically extracts deep features from the image through multi-layer convolution and pooling operations; these features characterize the semantic differences between the human body and the background. The last layer of the DCNN model (usually a convolutional or fully connected layer followed by upsampling) classifies each pixel in the image, determining whether it belongs to the "human target" or the "background." The final output is a segmentation mask of the same size as the input image, where the pixel region containing the target patient is marked as foreground (e.g., a value of 1), and the remaining background regions are marked as background (e.g., a value of 0). Using the segmentation mask, a pixel-by-pixel masking operation is performed on the original video image. Only the pixel regions marked as "human target" in the mask are retained, and all background pixels are removed, thus generating a "target region image" containing only the target patient's body region. For each frame of the processed "target region image," the timestamp information of its corresponding original video frame is associated to ensure accurate subsequent temporal analysis.

[0074] This method utilizes comprehensive monitoring to ensure the continuity and integrity of patient behavior data, and significantly improves the efficiency and targeting of data processing by automatically removing invalid video segments. The deep learning-based target segmentation algorithm can accurately identify and separate the patient's target area, reduce background interference, and enhance the reliability of subsequent behavioral feature extraction. At the same time, the timestamp marking provides a foundation for time series analysis, supporting the accurate tracking and analysis of the dynamic process of delirium.

[0075] Specifically, behavioral agitation of the target patient is quantified based on images of multiple target regions, including:

[0076] The Harris corner detection method is used to extract multiple key points of the target region image in each frame of video image, and the Harris response value corresponding to multiple key points is calculated. The key points include the head, torso and limbs of the target patient.

[0077] Optical flow is used to match key points of a target region in consecutive frames of video images to obtain multiple matching pairs and multiple motion vectors for the matching pairs.

[0078] The motion speed of the target patient in adjacent video frames is calculated based on multiple motion vectors, multiple Harris response values, and the consistency of the motion directions corresponding to the multiple motion vectors.

[0079] The number of connected components is calculated from multiple target region images, and the range of motion of the target patient is calculated from the positional changes of the target regions and the number of connected components.

[0080] Behavioral agitation is calculated based on the speed and amplitude of the movement.

[0081] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method first extracts key points such as the head, torso, and limbs from each frame of the target region image using the Harris corner detection method and obtains their response values. Then, it uses optical flow to match key points between consecutive frames to obtain matching pairs and their movement vectors. Based on the consistency of movement vectors, response values, and movement directions, the patient's movement speed between adjacent frames is calculated. At the same time, the movement amplitude is calculated by analyzing the number of connected components and the change in the position of the target region. Finally, the patient's behavioral agitation is quantified by combining movement speed and movement amplitude.

[0082] Harris corner detection is used to obtain multiple key points within the target area of ​​each frame of video image, as well as the corresponding Harris response value for each key point. The larger the Harris response value, the more important the key point is.

[0083] Key points generally include: head key points, torso key points, limb key points, hand and foot key points. These key points play an important role in human motion analysis and motion capture.

[0084] With a threshold of 10, within the time period of the w-th valid video segment, the 10 moments closest to moment t and moment t itself are obtained to form the reference time period for moment t.

[0085] During the reference time period at time t, optical flow (a known technique) is used to obtain all keypoints within the target region of the nth frame video image and their matching results with all keypoints within the target region of the (n+1)th frame video image, resulting in multiple successfully matched pairs. Each matching pair includes one keypoint from the nth frame and one keypoint from the (n+1)th frame.

[0086] Obtain the distance and direction from the position coordinates of the key point in the nth frame to the position coordinates of the key point in the (n+1)th frame in each matching pair, and construct the movement vector for each matching pair.

[0087] Then the movement speed of the target elderly patient corresponding to the nth frame and the (n+1)th frame. The calculation formula is:

[0088]

[0089] Where I is the number of matching pairs. Let Euclidean distance be the position coordinates of two keypoints in the i-th matching pair. The sum of the Harris response values ​​of the two keypoints in the i-th matching pair. This is the sum of the Harris response values ​​for all keypoints in all matching pairs. is the mean of the angle between the movement vectors of any two matching pairs.

[0090] In this method, a coordinate system is constructed with the bottom left corner of each video frame as the origin, the horizontal axis pointing to the right as the x-axis, and the vertical axis pointing upwards as the y-axis. The position coordinates of each pixel in each video frame are obtained within this coordinate system. Specifically, the more important the key point, the greater the distance it moves per unit of time (e.g., sudden hand or foot tremors, repeated hand rubbing, clapping in the air, limb twitching, etc., with aimless and rapid movements), and the more disordered and chaotic the movement direction of each key point, the better. The larger the size, the faster the target elderly patient's movement speed.

[0091] Within the reference time period at time t, the average motion velocity of the target elderly patient in all adjacent frames is obtained and denoted as the motion velocity of the target elderly patient at time t within the w-th valid video segment. .

[0092] Within the reference time period at time t, obtain the intersection region of the position coordinates of the target region in the nth frame video image and the target region in the (n+1)th frame video image. and the union of position coordinates , obtain The remaining coordinate region is then used to determine the number of connected components formed by adjacent coordinates. This reflects the number of body parts that were altered in this action.

[0093] Then The magnitude of the change in the range of motion of the target elderly patient corresponding to the nth frame and the (n+1)th frame. The mean value of the change in the amplitude of the target elderly patient's movement across all adjacent frames is obtained and used as the change in the amplitude of the target elderly patient's movement at time t within the w-th valid video segment. .

[0094] Where norm is a linear normalization function. This represents the magnitude of the change in the position coordinates of the target region from frame n to frame (n+1). The larger the value, the greater the change in the range of motion of the target elderly patient. Except for S2, which is a normalization operation. That is, the greater the range of motion of the target elderly patient, and the more body parts affected by the change in motion, the greater the change in the range of motion of the target elderly patient.

[0095] by As the target elderly patient's behavioral agitation at time t within the w-th effective video segment .

[0096] This method integrates keypoint response and optical flow tracking, which can accurately capture subtle movements of various parts of the patient's body and improve the accuracy of movement speed calculation. By combining connected component analysis and position change assessment of movement amplitude, the quantification of agitated behavior becomes more comprehensive and objective. Through the comprehensive calculation of multi-dimensional features (velocity, amplitude, and directional consistency), the robustness of behavioral agitation assessment is enhanced, providing a reliable and sensitive basis for subsequent delirium identification based on movement characteristics.

[0097] Specifically, the detailed changes in motion of the target patient are calculated based on images of multiple target regions, including:

[0098] Identify multiple key points of failed matching, calculate the number and importance of multiple key points of failed matching, and calculate the amount of change in detailed actions based on the number and importance of multiple key points of failed matching.

[0099] Among them, several key points of failed matching reflected the target patient's facial expressions and subtle limb tremors.

[0100] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. This method identifies key points that fail during optical flow matching and counts their number, while assessing the importance of these key points (usually corresponding to facial expressions and subtle limb tremors). Based on the number and importance of the key points that fail to match, the method calculates the amount of detail movement change that reflects the patient's subtle and irregular movement changes.

[0101] In addition to the significant changes mentioned above, postoperative delirium in elderly patients may also involve some subtle changes, such as changes in facial expressions (emotional fluctuations such as anxiety, fear, and anger), involuntary tremors or shaking of limbs, flailing of the air, limb twitching, and grasping of bed sheets. Therefore, further analysis of the subtle changes in the target area in each frame of video image is required.

[0102] For the reference time period at time t, among all key points in the target region of the nth frame video image and all key points in the target region of the (n+1)th frame video image, obtain the key points in the target region of the nth and (n+1)th frames video images that failed to match.

[0103] Keypoints may include the forehead, hairline, eyebrows, eyes, bridge of the nose, tip of the nose, corners of the mouth, chin, shoulders, chest, waist, abdomen, arms (shoulder joints, elbow joints, wrist joints), legs (hip joints, knee joints, ankle joints), fingertips, center of the palm, toes, and center of the foot. Therefore, these failed matches indicate that this important detail information has changed significantly or even been lost between frame n and frame n+1. Examples include rapid and drastic changes in facial expressions or limb twitches.

[0104] Within the reference time period at time t, obtain the number of all failed-matching keypoints within the target region in the video images of frame n and frame (n+1). and the sum of Harris response values ​​for all key points where matching failed. Obtain the number of all keypoints within the target region in the video images of frame n and frame (n+1). and the sum of Harris response values ​​for all key points. , obtain Let be the amount of detail change between frame n and frame (n+1). In other words, the more important key points change significantly, the greater the change in detail between adjacent frames.

[0105] Within the reference time period at time t, the mean of the detail changes corresponding to all adjacent frames is obtained, and denoted as the detail change of the target elderly patient at time t within the w-th valid video segment. .

[0106] This method can effectively capture subtle non-rigid movements that are easily overlooked in traditional motion analysis, such as facial muscle twitching or finger tremors, which are early subtle signs of delirium. By quantifying the matching failure points and their importance, it transforms the difficult-to-track detailed movements into calculable indicators, enhancing the sensitivity of identifying delirium-specific manifestations and providing an important supplementary dimension for comprehensively assessing changes in patient behavior.

[0107] Specifically, the severity of postoperative delirium in target patients was calculated based on the degree of behavioral agitation and changes in detail of motor function, including:

[0108] The mean values ​​of behavioral agitation and corresponding changes in detailed movements at each moment were calculated to obtain the degree of postoperative delirium.

[0109] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. This method directly generates the degree value of postoperative delirium at that time by averaging the behavioral agitation value and the change in detail of movement calculated at the same time, thereby obtaining multiple degree value sequences that change over time.

[0110] Obtain the behavioral agitation of the target elderly patient at time t within the w-th valid video segment. Changes in detailed motor skills in target elderly patients The mean value is denoted as the degree of postoperative delirium in the target elderly patient at time t within the w-th effective video segment.

[0111] The more agitated the target elderly patient was, and the greater the changes in subtle movements, the more pronounced the postoperative delirium. Similarly, the degree of postoperative delirium in the target elderly patient was obtained for each moment within each valid video segment.

[0112] This method is computationally simple and efficient, effectively integrating the behavioral agitation reflecting macroscopic movement with the detailed changes in movement reflecting microscopic changes, resulting in a comprehensive single-dimensional assessment index. This fusion approach takes into account both the overall activity level and local specificity of behavior, which helps to more evenly characterize the complex clinical manifestations of delirium and lays a clear data foundation for subsequent construction of continuous curves and trend analysis.

[0113] Specifically, the degree of postoperative delirium was fitted to a postoperative delirium performance curve, and multiple curve segments showing the occurrence of postoperative delirium were extracted from the postoperative delirium performance curve, including:

[0114] The degree of delirium at all times was fitted into a postoperative delirium curve in chronological order.

[0115] A preset threshold for the degree of delirium was set. Multiple postoperative delirium levels were compared with the threshold. All postoperative delirium levels exceeding the threshold were extracted from the postoperative delirium curve as the curve segments in which postoperative delirium occurred.

[0116] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method first connects the degree of postoperative delirium at all times in chronological order and fits to generate a continuous postoperative delirium performance curve. Then, by setting a delirium performance degree threshold, the time periods corresponding to all degree values ​​exceeding the threshold on the curve are identified and extracted. These time periods are defined as the curve segments where postoperative delirium occurs.

[0117] For the postoperative delirium levels of the target elderly patients at all times within all valid video segments, curve fitting is performed using the least squares method, with time as the horizontal axis and the postoperative delirium levels of the target elderly patients as the vertical axis (with values ​​ranging from 0 to 1), to obtain the curve of the postoperative delirium levels of the target elderly patients changing over time, which is denoted as the postoperative delirium curve.

[0118] The delirium severity threshold was set to 0.4. In the postoperative delirium performance curve, data points showing a delirium severity greater than 0.4 in the target elderly patients were recorded as target data points. The curve segment formed by adjacent target data points was recorded as the postoperative delirium occurrence curve segment. This resulted in multiple postoperative delirium occurrence curve segments.

[0119] This method achieves an intuitive, visual, and continuous display of the patient's delirium state through curve fitting, making it easy to observe its dynamic fluctuations. The threshold-based curve segment extraction method is simple and effective, and can automatically identify significant time periods when delirium symptoms may occur, providing a clear target range for subsequent focused analysis of the severity, frequency, and duration of symptoms, thus improving the efficiency and relevance of the analysis.

[0120] Specifically, the severity of multiple postoperative delirium symptoms was calculated based on the curve segments showing multiple postoperative delirium symptoms, including:

[0121] Calculate the skewness of the curve segments showing multiple postoperative delirium phenomena;

[0122] Calculate the cumulative value of the severity of multiple postoperative delirium symptoms along the curve segments;

[0123] The severity of postoperative delirium is obtained by multiplying multiple skewnesses by the cumulative values ​​of the manifestations corresponding to the multiple skewnesses.

[0124] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method calculates the skewness (reflecting the asymmetry of symptom intensity distribution) and the cumulative value of the manifestation severity (reflecting the total symptom load) of each identified postoperative delirium phenomenon curve segment, and multiplies the skewness of the same curve segment with its cumulative value to obtain multiple numerical values ​​that quantitatively characterize the severity of delirium manifestation at each time period.

[0125] Obtain the normalized value of the skewness of the curve segment where each postoperative delirium phenomenon occurs. Then, obtain the sum of the postoperative delirium severity of the target elderly patients for all data points within each curve segment where postoperative delirium occurs. , obtain and The product of these values ​​is used to represent the severity of postoperative delirium at each segment of the curve where delirium occurs. .

[0126] Skewness is an indicator of the symmetry of data distribution, and its calculation is a well-known technique. The skewness of all peak segments is normalized using the min-max normalization method.

[0127] Figure 3 A skewness diagram illustrating a behavioral pattern recognition-based early warning method for postoperative delirium in elderly patients provided in this application, as shown in the diagram. Figure 3 The diagram shows the skewness. Positive skewness indicates a skewness greater than 0, i.e. The larger the value, the more pronounced the postoperative delirium. Following a chronological order, the curve segment showing a rapid progression from normal to severe delirium within a short period, followed by a longer recovery time from severe delirium back to normal, indicates a rapid onset of postoperative delirium and a longer recovery period, meaning the delirium is more severe. The larger the value, the longer the postoperative delirium lasts, and the more severe the delirium.

[0128] In chronological order, all postoperative delirium occurrence curve segments were obtained, forming a postoperative delirium occurrence curve segment sequence. The severity of postoperative delirium manifestations for each of these curve segments was then obtained, forming a postoperative delirium severity sequence.

[0129] Specifically, the aggravation factor of postoperative delirium was calculated based on multiple curve segments of postoperative delirium and multiple severity levels of postoperative delirium, including:

[0130] The first slope was obtained by linear fitting of the severity of multiple postoperative delirium manifestations;

[0131] Clustering the time intervals of multiple postoperative delirium occurrence curve segments to obtain the duration of multiple delirium episodes, and fitting multiple durations to obtain a second slope.

[0132] The aggravation factor of postoperative delirium was calculated based on the first slope and the second slope.

[0133] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method first performs time-series linear fitting on the severity of multiple postoperative delirium manifestations to obtain a first slope reflecting the trend of severity change over time. At the same time, cluster analysis is performed on the time intervals between the segments of the delirium phenomenon to extract typical attack duration patterns, and a second slope is obtained by linear fitting on the duration. Finally, the aggravation factor of postoperative delirium manifestation is calculated by combining the first slope and the second slope.

[0134] Using the starting time of each postoperative delirium occurrence curve segment as the timestamp of the postoperative delirium severity for each segment, the least squares method was used to perform linear fitting on the postoperative delirium severity sequence, and the slope of the first fitted line was obtained. .

[0135] In the first fitted straight line, the vertical axis represents the severity of postoperative delirium, and the horizontal axis represents the timestamp corresponding to the severity of postoperative delirium.

[0136] Based on the sequence of postoperative delirium occurrence curve segments, the time interval between the postoperative delirium occurrence curve segments is used as the clustering factor. The K-means clustering algorithm is used to divide the sequence of postoperative delirium occurrence curve segments into multiple sequence segments.

[0137] In particular, a small time interval between adjacent postoperative delirium curve segments within each sequence segment indicates that postoperative delirium occurs continuously.

[0138] Obtain the total duration of all postoperative delirium curve segments in each sequence segment, and record it as the delirium duration of each sequence segment. In chronological order, construct a delirium duration sequence based on the delirium duration of all sequence segments.

[0139] Using the start time of the curve segment in which the first postoperative delirium phenomenon occurs within each sequence segment as the timestamp of the delirium duration for each sequence segment, the least squares method is used to fit a straight line to the delirium duration sequence, and the slope of the second fitted line is obtained. .

[0140] In the second fitted line, the vertical axis represents the duration of delirium, and the horizontal axis represents the timestamp corresponding to the duration of delirium.

[0141] Will This was recorded as a factor aggravating postoperative delirium in the target elderly patients. Where norm is a linear normalization function.

[0142] The larger K1 is, the more the severity of postoperative delirium gradually increases over time. The larger K2 is, the longer the duration of concentrated postoperative delirium, meaning that postoperative delirium occurs more frequently. This indicates that the postoperative delirium in the target elderly patient is gradually worsening.

[0143] This method comprehensively assesses disease progression from two dynamic dimensions: "symptom intensity evolution trend" and "attack time pattern evolution trend." It can quantify whether delirium tends to become more severe, more frequent, or more persistent. By extracting typical attack patterns through cluster analysis, it enhances the ability to discover patterns in fluctuating data. This allows the aggravation factor to not only reflect changes in historical severity but also reveal the potential evolution of attack duration, thus providing a more forward-looking and comprehensive assessment indicator for the risk of delirium progression.

[0144] Specifically, the overall severity of postoperative delirium was calculated based on multiple postoperative delirium severity and aggravating factors, including:

[0145] The total severity of postoperative delirium is obtained by summing the severity of multiple postoperative delirium symptoms.

[0146] The overall severity of postoperative delirium was obtained by multiplying the total severity of postoperative delirium by the aggravation factor and normalizing the result.

[0147] This application provides a method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition. The method first sums up the severity of multiple postoperative delirium manifestations calculated at different time periods to obtain the total severity of postoperative delirium manifestations, which reflects the overall symptom load. Then, this total severity is multiplied by a postoperative delirium manifestation aggravation factor that can characterize the dynamic deterioration trend of the condition, and the product result is normalized to finally obtain the comprehensive severity of postoperative delirium.

[0148] Obtain the sum of the severity of postoperative delirium manifestations for all segments of the curve showing postoperative delirium. ,by The overall severity of postoperative delirium in the target elderly patient at the current moment is recorded. .

[0149] In other words, if the severity of postoperative delirium in the target elderly patient is relatively high and gradually worsens during video monitoring, then the overall severity of postoperative delirium in the target elderly patient at the current moment is greater.

[0150] This method organically combines static cumulative severity with dynamic development trend factors, so that the comprehensive assessment index not only includes the overall severity of historical episodes, but also incorporates risk information on the possibility of the condition worsening. Through normalization, the standardization and comparability of the results are ensured, so that the comprehensive severity can more comprehensively and proactively reflect the overall clinical risk level of postoperative delirium in patients, providing a more in-depth quantitative basis for early warning decisions.

[0151] Thresholds of 0.3, 0.6, and 0.9 are set. If the overall severity Q of postoperative delirium in the target elderly patient is less than 0.3 at the current moment, the target elderly patient is considered to be currently normal.

[0152] when A P-value greater than or equal to 0.3 and less than 0.7 indicates that the target elderly patient is currently in a low-risk warning stage. The patient may exhibit mild behavioral abnormalities or inattention, prompting nurses to closely monitor the patient, ensure a quiet environment with adequate lighting, and avoid excessive stimulation. Simultaneously, the patient's pain control should be assessed, and potential causative medications such as anticholinergic drugs, benzodiazepines, and opioids should be identified. Good sleep quality should also be ensured for the patient.

[0153] When Q is greater than or equal to 0.7 and less than 0.9, the target elderly patient is considered to be in the medium-risk warning stage, and the patient may experience significant fluctuations in consciousness, disorientation, or mood swings. Nurses should further assess the patient's cognitive function, such as using the Mini-Mental State Examination (MMSE), and strengthen psychological care, encouraging family members to participate in the patient's care and provide emotional support. In addition, the patient's electrolytes, infections, and other medical complications should be closely monitored, and potential triggers should be addressed promptly.

[0154] When Q is greater than or equal to 0.9, the target elderly patient is considered to be in a high-risk warning stage. The patient may exhibit severe altered consciousness, hallucinations, or delusions. The nurse should immediately notify the attending physician, administer the prescribed medication, and continuously monitor the treatment effect. Simultaneously, the use of restraints or other restrictive measures should be avoided unless absolutely necessary to reduce the patient's anxiety and discomfort.

[0155] Figure 4 A schematic diagram of the structure of a postoperative delirium early warning system for elderly patients based on behavioral pattern recognition provided in this application is shown below. Figure 4 As shown in this embodiment, a postoperative delirium early warning system for elderly patients based on behavioral pattern recognition is provided. The system includes:

[0156] The data acquisition and preprocessing unit is used to acquire videos of the target patient's daily activities and obtain images of the target region in each frame of the daily activity videos.

[0157] The feature extraction and postoperative delirium severity calculation unit is connected to the data acquisition and preprocessing unit. The feature extraction and postoperative delirium severity calculation unit is used to quantify the behavioral agitation of the target patient based on multiple target region images and to calculate the amount of change in the detailed movements of the target patient. The postoperative delirium severity of the target patient is calculated based on the behavioral agitation and the amount of change in the detailed movements.

[0158] The temporal analysis and comprehensive severity assessment unit is connected to the feature extraction and postoperative delirium severity calculation unit. The temporal analysis and comprehensive severity assessment unit is used to fit the postoperative delirium severity into a postoperative delirium performance curve, extract multiple curve segments of postoperative delirium occurrence from the postoperative delirium performance curve, calculate multiple postoperative delirium severity based on multiple postoperative delirium occurrence curve segments, calculate postoperative delirium aggravation factor based on multiple postoperative delirium occurrence curve segments and multiple postoperative delirium severity, and calculate the comprehensive severity of postoperative delirium based on multiple postoperative delirium severity and postoperative delirium aggravation factor.

[0159] The risk grading and early warning response unit is connected to the time series analysis and comprehensive severity assessment unit. The risk grading and early warning response unit is used to preset severity thresholds and compare the comprehensive severity of postoperative delirium with the severity threshold to obtain the corresponding severity warning.

[0160] This application provides a postoperative delirium early warning system for elderly patients based on behavioral pattern recognition. The system acquires and extracts target area images from patient videos through a data acquisition and preprocessing unit. A feature extraction and postoperative delirium severity calculation unit quantifies the amount of behavioral agitation and changes in detailed movements and fuses them into the postoperative delirium severity. Subsequently, a time series analysis and comprehensive severity assessment unit fits the severity into a curve, extracts the epidural curve segment, and calculates the severity and aggravation factors to obtain the comprehensive severity. Finally, a risk grading and early warning response unit compares it with a preset threshold to achieve automated graded early warning.

[0161] The system constructs a fully automated closed loop from data acquisition to early warning response. Each unit has a clear division of labor and close connection, realizing multi-level and dynamic assessment of postoperative delirium. By integrating microscopic details of actions with macroscopic behavioral agitation and combining time-series trend analysis, the accuracy and foresight of early warning are significantly improved. The systematic structure also enhances the reliability, scalability and convenience of clinical deployment of the solution.

Claims

1. A method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition, characterized in that, The early warning method includes: Collect videos of the target patient's daily activities and obtain images of the target region in each frame of the daily activity videos; The behavioral agitation of the target patient is quantified based on multiple images of the target region, and the amount of change in the detailed movements of the target patient is calculated. The degree of postoperative delirium of the target patient is calculated based on the behavioral agitation and the amount of change in the detailed movements. The degree of postoperative delirium was fitted to a postoperative delirium performance curve, and multiple curve segments in which postoperative delirium occurred were extracted from the postoperative delirium performance curve. Multiple postoperative delirium severity was calculated based on the multiple postoperative delirium performance curve segments. Postoperative delirium aggravation factor was calculated based on the multiple postoperative delirium performance curve segments and the multiple postoperative delirium performance severity. The overall severity of postoperative delirium was calculated based on the multiple postoperative delirium performance severity and the postoperative delirium aggravation factor. A severity threshold is preset, and the overall severity of postoperative delirium is compared with the severity threshold to obtain a corresponding severity warning.

2. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 1, characterized in that, The process of collecting videos of the target patient's daily activities and obtaining target region images for each frame of the videos includes: The daily activities of the target patient are captured without blind spots using the monitoring equipment in the ward where the target patient is located; Invalid video segments from the daily activity videos that do not contain the target patient are removed, and valid video segments containing the target patient are retained. The target region image in each frame of the effective video segment is obtained using a human target segmentation algorithm based on a deep convolutional neural network, wherein each frame of the video image corresponds to a timestamp.

3. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 1, characterized in that, The step of quantifying the behavioral agitation of the target patient based on multiple target region images includes: The Harris corner detection method is used to extract multiple key points of the target region image in each frame of the video image, and the Harris response value corresponding to the multiple key points is calculated. The key points include the head, torso and limbs of the target patient. Optical flow is used to match key points of the target region image in consecutive frames of the video image to obtain multiple matching pairs and multiple movement vectors of the multiple matching pairs; The motion speed of the target patient in adjacent frames of the video image is calculated based on the consistency of multiple motion vectors, multiple Harris response values, and the motion directions corresponding to multiple motion vectors. The number of connected components is calculated using multiple images of the target region, and the amplitude of the target patient's movement is calculated using the positional changes of the target region and the number of connected components. The agitation of the behavior is calculated based on the speed and amplitude of the movement.

4. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 2, characterized in that, The step of calculating the detailed movement changes of the target patient based on multiple target region images includes: Identify multiple key points that fail to match, calculate the number and importance of the multiple key points that fail to match, and calculate the change in the detailed action based on the number and importance of the multiple key points that fail to match. Among them, the key points where multiple matches failed reflect the target patient's facial expressions and subtle limb tremors.

5. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 1, characterized in that, The calculation of the postoperative delirium severity in the target patient based on the behavioral agitation and the amount of change in detailed motor actions includes: The mean of the behavioral agitation and the corresponding change in the detailed action at each moment is calculated to obtain the degree of postoperative delirium.

6. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 1, characterized in that, The process involves fitting the degree of postoperative delirium to a postoperative delirium performance curve, and extracting multiple curve segments from the postoperative delirium performance curve that indicate the occurrence of postoperative delirium, including: The degree of delirium at all times was fitted in chronological order to form the postoperative delirium performance curve; A preset delirium severity threshold is set, and multiple postoperative delirium severity levels are compared with the delirium severity threshold. All postoperative delirium severity levels in the postoperative delirium severity curve that are greater than the delirium severity threshold are extracted as the curve segments in which the postoperative delirium phenomenon occurs.

7. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 1, characterized in that, The calculation of the severity of multiple postoperative delirium manifestations based on multiple curve segments of the postoperative delirium phenomenon includes: Calculate the skewness of the curve segments in which multiple postoperative delirium phenomena occur; Calculate the cumulative value of the manifestation degree of multiple curve segments of the postoperative delirium phenomenon; The severity of postoperative delirium is obtained by multiplying the multiple skewnesses by the cumulative value of the manifestation corresponding to the multiple skewnesses.

8. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 1, characterized in that, The calculation of the postoperative delirium aggravation factor based on multiple curve segments of postoperative delirium occurrence and multiple postoperative delirium severity includes: The first slope was obtained by linear fitting of the severity of multiple postoperative delirium manifestations; Cluster the time intervals of the curve segments of multiple postoperative delirium phenomena to obtain the duration of multiple delirium episodes, and fit multiple durations to obtain a second slope; The aggravation factor of postoperative delirium is calculated based on the first slope and the second slope.

9. The method for early warning of postoperative delirium in elderly patients based on behavioral pattern recognition according to claim 1, characterized in that, The calculation of the overall severity of postoperative delirium based on multiple postoperative delirium severity manifestations and postoperative delirium aggravation factors includes: The total severity of postoperative delirium is obtained by summing the severity of multiple postoperative delirium symptoms. The overall severity of postoperative delirium is obtained by multiplying the total severity of postoperative delirium by the aggravation factor and normalizing the result.

10. A postoperative delirium early warning system for elderly patients based on behavioral pattern recognition, wherein the system applies the method according to any one of claims 1 to 9, and the system comprises: A data acquisition and preprocessing unit is used to acquire the daily activity video of the target patient and obtain the target region image of each frame in the daily activity video; The feature extraction and postoperative delirium severity calculation unit is connected to the data acquisition and preprocessing unit. The feature extraction and postoperative delirium severity calculation unit is used to quantify the behavioral agitation of the target patient based on multiple target region images and to calculate the amount of change in the detailed movements of the target patient. The postoperative delirium severity of the target patient is calculated based on the behavioral agitation and the amount of change in the detailed movements. A time-series analysis and comprehensive severity assessment unit is connected to the feature extraction and postoperative delirium severity calculation unit. The time-series analysis and comprehensive severity assessment unit is used to fit the postoperative delirium severity to the postoperative delirium performance curve, extract multiple curve segments of the postoperative delirium phenomenon occurrence from the postoperative delirium performance curve, calculate multiple postoperative delirium severity based on the multiple postoperative delirium phenomenon occurrence curve segments, calculate the postoperative delirium aggravation factor based on the multiple postoperative delirium phenomenon occurrence curve segments and the multiple postoperative delirium severity, and calculate the comprehensive severity of postoperative delirium based on the multiple postoperative delirium severity and the postoperative delirium aggravation factor. The risk grading and early warning response unit is connected to the time series analysis and comprehensive severity assessment unit. The risk grading and early warning response unit is used to preset the severity threshold and compare the comprehensive severity of postoperative delirium with the severity threshold to obtain the corresponding severity warning.

Images