Methods and Systems for Monitoring Abnormal Behaviors in Critically Ill Patients Based on Spatiotemporal Imagery
By using a spatiotemporal image-based method, we can identify the hidden behaviors of critically ill patients and construct a dynamic mechanical relationship model. This solves the problems of missed detection and false alarms of behaviors under bedding in existing technologies, and enables accurate identification and risk warning of hidden contact constraints.
Patent Information
- Application Number
- CN202611266304.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for monitoring the behavior of critically ill patients are difficult to identify grasping, pressing, entanglement or hooking behaviors under the bedding in real time, and are easily affected by the rebound of the bedding, vibration of the hospital bed and free swing of the tubing, leading to missed detections and false alarms.
By acquiring continuous spatiotemporal images of patients, bedding, and medical tubing, the system identifies limb joints and tubing nodes, establishes a baseline perturbation response template, matches real-time response sequences, constructs a dynamic mechanical relationship model, predicts and generates graded early warning commands, and enables the identification and risk warning of hidden contact constraints.
Hidden behaviors can be identified without directly observing under the bedding, improving the accuracy of abnormal behavior identification, reducing false alarm rates, and enabling early warning and precise handling of delayed dangers.
Smart Images

Figure CN123120214A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical monitoring and image processing technology, and in particular to a method and system for monitoring abnormal behavior of critically ill patients based on spatiotemporal images. Background Technology
[0002] Critically ill patients are typically connected to ventilator tubing, intravenous lines, and monitoring cables during monitoring. Due to factors such as altered consciousness, unconscious limb movements, and turning over, patients may exhibit abnormal behaviors such as grasping, pressing, tangling, or hooking onto medical tubing. When such behaviors are obscured by bedding, nursing staff and regular video monitoring may find them difficult to detect in time. This can easily lead to reduced mobility of medical tubing, misalignment of tubing fixation points, or localized bending during subsequent head rotations, limb repositioning, and adjustments to bed posture.
[0003] Current monitoring of critically ill patients' behavior mainly relies on manual rounds, bedside cameras, limb posture recognition, or medical tubing displacement detection. Manual rounds are difficult to conduct continuous monitoring; bedside cameras and posture recognition can usually only identify visible limb movements and cannot determine whether there is grasping, pressing, entanglement, or hooking under the bedding; medical tubing displacement detection usually requires the tubing to have moved significantly before triggering an alarm, making it difficult to detect hidden contact constraints that have not yet caused direct danger.
[0004] Furthermore, existing methods often use fixed thresholds to judge patient movements or medical tubing displacement, failing to establish individualized benchmarks based on patient posture, bedding coverage, medical tubing layout, and local bed response. These methods are easily affected by bedding rebound, bed vibration, and free swaying of tubing. Therefore, this invention proposes a method and system for monitoring abnormal behavior in critically ill patients based on spatiotemporal imaging. Summary of the Invention
[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a method and system for monitoring abnormal behaviors of critically ill patients based on spatiotemporal imaging, thereby resolving the technical problems mentioned in the background section.
[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring abnormal behavior in critically ill patients based on spatiotemporal imaging includes the following steps: S1. Acquire continuous spatiotemporal images of the patient, bedding, bed, and medical tubing; identify joints of the patient's limbs and nodes of the medical tubing; determine the target area of the bed; perform predetermined micro-displacement; and establish a baseline perturbation response template. S2. Match the target reference perturbation response template, perform a predetermined micro-displacement, obtain the real-time perturbation response sequence and calculate the response difference, identify gripping, pressing, entanglement or hooking type hidden contact constraints, and form a set of hidden contact constraints. S3. Based on the hidden contact constraint set, the patient's limb joints, the medical tubing nodes and their fixed points and equipment connection ends, establish a spatial mechanical node set, determine the node connections, the remaining active length of the medical tubing and the transmission path, and construct a dynamic mechanical relationship model. S4. Generate a set of events to be simulated based on the patient's motion records, bedside instructions to be executed, and medical equipment operating parameters. Input the dynamic mechanical relationship model to obtain event response prediction results and control response prediction results, and determine the delayed danger coupling results. S5. Determine the risk level based on the delayed danger coupling result, generate graded early warning instructions and pre-authorized equipment control instructions; after handling, execute the predetermined micro-displacement again, and after confirming that the hidden contact constraint has disappeared and there is no delayed danger coupling, cancel the corresponding instructions and update the benchmark micro-perturbation response template and dynamic mechanical relationship model.
[0007] A spatiotemporal image-based system for monitoring abnormal behaviors in critically ill patients includes: The spatiotemporal image and benchmark establishment module is used to acquire continuous spatiotemporal images of patients, bedding, hospital beds and medical tubing, and establish benchmark perturbation response templates; The hidden contact constraint identification module is used to match the target reference perturbation response template, obtain the real-time perturbation response sequence, and form a hidden contact constraint set. The dynamic mechanical relationship modeling module is used to establish a set of spatial mechanical nodes based on the hidden contact constraint set, patient limb joints, medical tubing nodes, tubing fixing points, bed boundaries and equipment connection ends, and to construct a dynamic mechanical relationship model. The delayed danger simulation module is used to generate a set of events to be simulated based on the patient's motion records, the instructions to be executed on the hospital bed, and the operating parameters of the medical equipment, and input the set of events to be simulated into the dynamic mechanical relationship model; The risk handling and de-escalation verification module is used to determine the risk level based on the delayed hazard coupling result, and generate graded early warning instructions and pre-authorized equipment control instructions.
[0008] The beneficial effects of this invention are as follows: This invention obtains the response differences between bedding and medical tubing by pre-defined micro-displacements in the target area of the hospital bed. It can identify hidden grasping, pressing, entanglement, and hooking behaviors without direct observation under the bedding, reducing missed detections caused by occlusion. A baseline micro-perturbation response template is established based on the patient's posture, medical tubing layout, bedding coverage, and the target area of the hospital bed. This template can distinguish abnormal constraint responses from bedding rebound, bed vibration, and free swaying of tubing, improving the accuracy of abnormal behavior identification.
[0009] This invention, through joint analysis of medical tubing node displacement, node movement sequence, deformation propagation delay, and recovery path, can determine the type, location, direction, and degree of hidden contact constraints, reducing false alarms caused by relying solely on limb position. By establishing the dynamic mechanical relationship between the patient's limb, hidden contact constraints, medical tubing, tubing fixation points, and equipment connection ends, the displacement transmission path can be traced, clarifying the causes of reduced medical tubing mobility, fixation point traction, and localized bending.
[0010] This invention, by comparing and extrapolating constrained and unconstrained states of patient movement, bed movements, and cyclical movements of medical equipment, can identify delayed dangerous couplings before actual traction or bending occurs, thus achieving early warning of risks. It implements tiered warnings based on risk level and prediction time, restricting only authorized non-treatment-essential movements; after intervention, micro-displacement verification confirms risk relief, avoiding misinterpretation of warnings, and continuously updating the monitoring template and dynamic mechanical relationship model. Attached Figure Description
[0011] Figure 1 This is a flowchart of the method for monitoring abnormal behavior of critically ill patients based on spatiotemporal images according to the present invention; Figure 2 This is a framework diagram of the critical patient abnormal behavior monitoring system based on spatiotemporal imaging according to the present invention. Detailed Implementation
[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0013] Example 1: As Figure 1 As shown, this embodiment provides a method for monitoring abnormal behavior in critically ill patients based on spatiotemporal images, including the following steps: S1. Acquire continuous spatiotemporal images of the patient, bedding, bed, and medical tubing; identify joints of the patient's limbs and nodes of the medical tubing; determine the target area of the bed; perform predetermined micro-displacement; and establish a baseline perturbation response template. S2. Match the target reference perturbation response template, perform a predetermined micro-displacement, obtain the real-time perturbation response sequence and calculate the response difference, identify gripping, pressing, entanglement or hooking type hidden contact constraints, and form a set of hidden contact constraints. S3. Based on the hidden contact constraint set, the patient's limb joints, the medical tubing nodes and their fixed points and equipment connection ends, establish a spatial mechanical node set, determine the node connections, the remaining active length of the medical tubing and the transmission path, and construct a dynamic mechanical relationship model. S4. Generate a set of events to be simulated based on the patient's motion records, bedside instructions to be executed, and medical equipment operating parameters. Input the dynamic mechanical relationship model to obtain event response prediction results and control response prediction results, and determine the delayed danger coupling results. S5. Determine the risk level based on the delayed danger coupling result, generate graded early warning instructions and pre-authorized equipment control instructions; after handling, execute the predetermined micro-displacement again, and after confirming that the hidden contact constraint has disappeared and there is no delayed danger coupling, cancel the corresponding instructions and update the benchmark micro-perturbation response template and dynamic mechanical relationship model.
[0014] S1 specifically includes the following sub-steps: S110. Image acquisition devices are installed above and to the side of the hospital bed to continuously acquire color images and depth data of critically ill patients, bedding, hospital bed, medical tubing, tubing fixing points and equipment connection ends according to the same system clock, and an acquisition timestamp is added to each frame of data.
[0015] The image acquisition device uses a depth camera, or two cameras with overlapping fields of view and dual-target positioning. Color images, depth data and acquisition timestamps are output by the image acquisition device. The position, tilt angle and drive status of each adjustable area of the bed are output by the bed controller. The actual displacement of the target area of the bed is output by the encoder of the corresponding drive unit.
[0016] The fixed mark on the left side of the head end of the bed board is taken as the origin of the coordinate system. The direction from the head end of the bed to the feet, the direction from the left side of the bed to the right side, and the direction perpendicular to the bed board upward are taken as the three coordinate axes of the bed space coordinate system.
[0017] At least three non-collinear visual calibration marks with known positions are set at the edge of the hospital bed. Coordinate transformation parameters are obtained based on the pixel coordinates of the visual calibration marks in the image and their actual coordinates in the hospital bed spatial coordinate system. The target position in the image acquisition device coordinate system is then transformed to the hospital bed spatial coordinate system according to the following formula: in, This represents the 3D coordinates of the target point in the coordinate system of the image acquisition device. This represents the 3D coordinates of the target point in the bed space coordinate system. This represents the rotation matrix from the image acquisition device coordinate system to the bed space coordinate system. The corresponding translation vector is represented; the rotation matrix and translation vector are obtained during the installation and calibration stage and are recalibrated after the position of the image acquisition device changes or the entire bed is moved.
[0018] Image recognition is performed on continuous spatiotemporal images. Specifically: a deep learning-based instance segmentation model (such as Mask R-CNN) is used to identify the patient's head contour, bedding coverage area, bed boundary, and equipment connection point; a human pose estimation network (such as OpenPose) is used to extract the patient's shoulder joints, chest region, upper limb joints, hand region, and lower limb joints; an edge detection operator (such as the Canny operator) combined with the Hessian matrix analysis algorithm is used to extract the bedding fold ridge line and local bulge areas; and a skeleton extraction and spline curve fitting algorithm is used to extract the medical pipeline centerline, medical pipeline nodes, and pipeline fixing points in the visible pipeline area.
[0019] Object association is performed based on the positional distance, contour overlap, motion direction, and appearance features of target objects in adjacent image frames; for medical pipelines, identity verification is further performed by combining centerline continuity, pipeline fixing points, and equipment connection ends, and continuous tracking identifiers are configured for each target object to generate a baseline object tracking sequence.
[0020] Independently drivable bed board units or air mattress units are selected as candidate areas. The target area for the bed is chosen when a candidate area simultaneously covers the patient's limb projection area and the location of medical tubing insertion, avoids the patient's head and neck weight-bearing area, and is not designated as a restricted area by nursing staff. For example, if the patient's right forearm and the first ventilator tubing are both located under the bedding and overlap with the projection area of the sixth air mattress unit, then the sixth air mattress unit is identified as the target area for the bed.
[0021] S120. Read the baseline object tracking sequence and determine whether the baseline perturbation acquisition conditions are met within a stable observation window of 3 s to 10 s. The baseline perturbation acquisition conditions are met when the displacement and velocity of the patient's target limb do not exceed the corresponding threshold, no medical staff are identified in the nursing area, the displacement of the tubing fixation point and the change in the length of the medical tubing centerline are within the natural fluctuation range of the stable observation window, and the bed drive unit is operating normally and there is no signal prohibiting bed movement.
[0022] The maximum allowable displacement for patient care is entered by the nursing staff based on the patient's current condition; the maximum displacement for bed detection is read by the bed controller; the minimum identifiable displacement of the image acquisition device is obtained through calibration testing; and the predetermined micro-displacement amplitude is determined according to the following formula: in, This indicates the actual micro-displacement amplitude of the target area of the hospital bed. This indicates the upper limit of the probe displacement that the bed drive unit is allowed to perform. This indicates the upper limit of permissible displacement for patient care; the actual micro-displacement amplitude is not less than the minimum identifiable displacement of the image acquisition device.
[0023] The target area of the controlled bed is controlled to perform one micro-lift / lower movement along a pre-set direction. The actual displacement, movement speed, and start, hold, and cancel timestamps fed back by the encoder of the drive unit are recorded. The ridge lines of bedding folds, localized raised areas, medical tubing nodes, and tubing fixation points are continuously tracked. The spatial coordinates of the bed, the actual displacement of the target area, and the corresponding timestamps of each target object at each sampling time are arranged in chronological order to form a single-sample baseline micro-perturbation response. This process is repeated 3 to 10 times while maintaining consistent patient posture, target area of the bed, medical tubing layout, and bedding coverage.
[0024] If, during execution, the system detects patient movement, entry of medical staff into the nursing area, displacement velocity of the tubing fixed point exceeding the fixed point velocity threshold, loss of continuous tracking marker, abnormality of the drive unit, or a signal prohibiting bed movement, the predetermined micro-displacement is immediately revoked, and this sample is marked as invalid. For example, if the patient's right hand displacement exceeds the active movement threshold during micro-lifting, the 6th air cushion unit is immediately returned to its original position, and this response is not included in the template construction. The remaining samples are then used to form a valid baseline perturbation response sample sequence.
[0025] S130. Taking the moment when the encoder detects the start of movement in the target area of the bed as the zero point of time, and the moment when the target area of the bed returns to its initial position as the end time of perturbation cancellation, each valid sample is divided into a stable segment, an execution segment, a hold segment, a cancellation segment, and a recovery segment, and resampled at a uniform sampling interval. When the proportion of missing image frames does not exceed the allowable proportion determined by the continuity verification experiment, the timestamps and three-dimensional spatial coordinates of the adjacent valid frames before and after the missing frame are extracted, and the missing coordinates are filled in using a time-based linear interpolation algorithm; when the proportion exceeds the allowable proportion, the corresponding sample is discarded.
[0026] The displacement of the target object is normalized by using the actual micro-displacement amplitude recorded by the encoder: in, This represents the 3D spatial coordinates of the i-th target object at time t. This indicates the moment before the perturbation is applied to the target object. 3D spatial coordinates This indicates the actual micro-displacement amplitude of the target area of the hospital bed. This represents the normalized displacement vector corresponding to the micro-displacement of a unit hospital bed.
[0027] The spatial expansion direction formed by the order in which each feature point of the bedding first reaches the effective displacement threshold is defined as the direction of bedding deformation propagation; the time difference between the start of movement of the target area of the hospital bed and the first effective displacement threshold of the target feature point is defined as the bedding deformation propagation delay; the order in which the medical pipeline nodes first reach the effective displacement threshold is defined as the node movement sequence; the time required for the target object to enter and continuously remain within the allowable error range of the initial position after the target area of the hospital bed is restored to its original position is defined as the recovery time; and the continuous spatial trajectory of the target object from the start of perturbation removal to the period of restoration to stability is defined as the recovery path.
[0028] The effective displacement threshold and the allowable error range of the initial position are both determined based on the mean of the natural fluctuations in the stable segment and three times the standard deviation.
[0029] The effective samples were grouped according to the patient's posture, the target area of the bed, the medical tubing markings, the tubing fixing points, and the bedding covering method. The mean, standard deviation, and allowable fluctuation range of each response feature were calculated to establish a baseline perturbation response template. The template's applicable conditions and the template itself were stored in the template library.
[0030] When changes occur in patient posture, medical tubing identification, tubing fixation points, bedding coverage, target area of the bed, or calibration status of the image acquisition device, the original template is stopped from being invoked and S110 to S130 are re-executed. The baseline perturbation response template is stored in the template library for template applicability matching in S210.
[0031] S2 specifically includes the following sub-steps: S210. During normal monitoring, continuous spatiotemporal images, bed controller data, drive unit encoder data, and template library are read according to a unified system timestamp. The patient's limb joints, hand area, bedding coverage area, medical tubing centerline, medical tubing nodes, tubing fixing points, equipment connection ends, and medical staff positions are derived from the image recognition results of the continuous spatiotemporal images; the bed posture, drive unit status, and prohibition of bed movement signals are derived from the bed controller; the actual displacement of the target area of the bed is derived from the drive unit encoder.
[0032] Based on the current patient posture, target area of the bed, medical pipeline markings, pipeline fixing points, equipment connection ends, bedding covering method, and bed tilt angle, a baseline perturbation response template that meets the applicable conditions is selected from the template library, and the selected template is determined as the target baseline perturbation response template.
[0033] The allowable range of bed tilt angle is based on the tilt angle fluctuation range recorded when the corresponding template was established; the overlap of the bedding coverage area is based on the ratio of the intersection area of the current coverage area and the template coverage area to the union area of the two, and its threshold is determined by the template verification test. If no applicable template exists, real-time perturbation is not performed, and the process returns to S110 to S130 to establish a new baseline perturbation response template.
[0034] When the target area of the hospital bed simultaneously covers the patient's limb projection area and the section of the medical tubing obscured by the bedding, and the medical tubing has identifiable entry and exit points and visible tubing sections continuously connected to the tubing's fixing points, the target area of the hospital bed is determined as the current detection area.
[0035] Within a stable observation window of 3 to 10 seconds, after confirming that the patient's target limb has not undergone excessive active movement, there are no medical staff in the nursing area, no excessive displacement of the tubing fixation points, and the drive unit is operating normally without any signals prohibiting bed movement, the target area of the bed is controlled to perform a predetermined micro-displacement identical to the target baseline perturbation response template. The 3D spatial coordinates of the bedding folds, ridges, local raised areas, medical tubing nodes, and tubing fixation points, as well as the actual encoder displacement and timestamps at each stage, are continuously recorded to form a real-time perturbation response sequence.
[0036] If, during execution, the patient moves actively, medical staff enter the nursing area, the displacement velocity of the tubing fixation point exceeds the fixation point velocity threshold, the target object is lost, or the drive unit malfunctions, the predetermined micro-displacement is immediately revoked, and the current sequence is marked as invalid. For example, if the patient changes from supine to right lateral decubitus, even if the first ventilator tubing still passes through the sixth air cushion unit, the template corresponding to the supine state must not be invoked.
[0037] S220. Taking the moment when the encoder detects the start of movement in the target area of the bed as the zero point of time, the real-time perturbation response sequence and the target reference perturbation response template are divided into execution segment, hold segment, cancel segment and recovery segment respectively, and time-aligned according to the same sampling interval; the spatial deviation caused by the slight shaking of the image acquisition device is corrected according to the fixed mark of the bed; the displacement is normalized by using the actual micro-displacement amplitude recorded by the encoder.
[0038] For the j-th response feature, the standardized variance is calculated according to the following formula: in, This represents the standardized variance of the j-th key response feature. This represents the numerical value of the j-th response feature in the real-time perturbation response sequence. This represents the mean value of the corresponding feature in the target baseline perturbation response template. This represents the standard deviation of the corresponding feature. This represents a positive number with the same dimensions as the j-th response feature, used to prevent a denominator of 0. When the standardized variance exceeds the variance threshold determined by template validation experiments, the response is considered to be outside the normal fluctuation range.
[0039] The time difference between the first effective displacement threshold reached by the same bedding feature point in the real-time response and the baseline response is defined as the bedding deformation propagation delay difference; the distance between the real-time displacement vector and the baseline displacement vector of the same medical pipeline node at the same normalized time is defined as the medical pipeline node displacement difference; the proportion of the number of reversed pairs formed by the real-time node movement order relative to the baseline node movement order to the total number of node pairs is defined as the node movement order difference; the average distance of spatial points obtained by pairing the real-time recovery path and the baseline recovery path at a unified sampling time, divided by the length of the baseline recovery path, is defined as the recovery path deviation degree, with a larger value indicating a more obvious deviation.
[0040] The aforementioned differences, pipeline fixed-point offset differences, and recovery time differences are combined to extract the corresponding medical pipeline identifiers, difference locations, duration of response deviations, and propagation directions established based on effective displacements. Finally, the ratio of each difference quantity to its corresponding difference threshold is defined as the degree of deviation, and these are combined into a real-time difference response feature set. For example, in the baseline state, medical pipeline nodes 1 to 4 move sequentially, while in the real-time response, the displacement of medical pipeline node 3 decreases and medical pipeline node 4 moves in the opposite direction before medical pipeline node 3. In this case, abnormal node displacement and abnormal node movement sequence are recorded near medical pipeline node 3.
[0041] S230. Read the real-time difference response feature set and aggregate medical tubing nodes that continuously exceed the difference threshold and are spatially adjacent into abnormal tubing segments. A hidden contact constraint is determined to exist only when the abnormal features are continuously maintained for no less than a predetermined number of frames, the drift of the abnormal position in consecutive frames does not exceed the position stability threshold, and it is supported by at least two types of response features; the predetermined number of frames and the position stability threshold are determined by image recognition stability test.
[0042] When the length of the abnormal tube segment is not greater than the grasping length threshold, the movement directions of the medical tube nodes on both sides of the abnormal location are opposite or the displacement propagation is interrupted, and the minimum distance from the abnormal location to the patient's hand projection boundary is not greater than the hand adjacency distance threshold, it is determined to be a grasping-type hidden contact constraint, and the medical tube node with the most obvious propagation interruption is determined to be the grasping position. When the displacement amplitude of at least two consecutive medical tube nodes decreases sequentially, the abnormal tube segment coincides with the pressure projection of the patient's limb and the recovery time increases, it is determined to be a holding-type hidden contact constraint, and the geometric center of the abnormal tube segment is determined to be the holding position.
[0043] When the movement order of medical tubing nodes is reversed, the change in curvature of the centerline exceeds the curvature change threshold, the movement directions of medical tubing nodes on both sides of the patient's limb are opposite, and the deviation of the recovery path after the perturbation is removed exceeds the corresponding threshold, it is determined to be a tangled hidden contact constraint, and the centerline position with the largest change in curvature is determined as the tangling position.
[0044] When an abnormal location is near a finger, bed rail, or bedding edge, and the medical tubing can move in the disengagement direction but is restricted in the tightening direction, and the difference in displacement response between the execution and retraction segments exceeds the directional asymmetry threshold, it is identified as a hook-type hidden contact constraint, and the medical tubing node with the largest difference in forward and reverse displacement is identified as the hook location. The aforementioned length threshold, distance threshold, curvature change threshold, and directional asymmetry threshold are all determined by pre-calibration tests under no-abnormal and corresponding constraint states. In one specific embodiment, the grasping length threshold is set to 80mm–120mm, the hand proximity distance threshold is set to 30mm–50mm, and the curvature change threshold is set to 0.05mm. -1 ~0.1 mm -1 The directional asymmetry threshold is set to 1.5 to 2.0 times the effective displacement threshold.
[0045] The main displacement direction restricted by the hidden contact constraint position is defined as the hidden contact constraint direction, which is determined based on the difference between the real-time displacement vector and the reference displacement vector. The restriction strength of the hidden contact constraint on the normal perturbation response is defined as the degree of hidden contact constraint, which is calculated by normalized weighting based on the difference in medical pipeline node displacement, the difference in the propagation delay of bedding deformation, the difference in node motion order, and the deviation of the recovery path. Each weight is determined by a pre-calibration test and the sum is 1.
[0046] When the same medical tubing simultaneously meets multiple constraints, multiple constraint records are output; if it cannot be separated, a composite hidden contact constraint is output. Each record includes the medical tubing identifier, constraint type, constraint location, constraint direction, constraint degree, and identification confidence level, forming a hidden contact constraint set and sending it to S310.
[0047] S3 specifically includes the following sub-steps: S310: Read the hidden contact constraint set output by S230, and obtain the current continuous spatiotemporal image, bed controller data and medical pipeline registration information according to the unified system timestamp.
[0048] Each record in the hidden contact constraint set includes medical tubing identification, constraint type, constraint location, constraint direction, constraint degree, and identification confidence level; patient limb joints, medical tubing centerline, medical tubing nodes, tubing fixing points, equipment connection ends, bed boundaries, bed rail boundaries, and bedding obstruction boundaries are derived from image recognition results of continuous spatiotemporal images; bed posture and adjustable area location are derived from the bed controller and drive unit encoder; medical tubing category, outer diameter, registered length, connection location, fixing method, minimum allowable bending radius, and medical tubing risk attributes are derived from medical tubing registration information entered by nursing staff or corresponding product parameters.
[0049] Spatial position units used to represent displacement input, displacement transmission, motion restriction, or displacement output are defined as mechanical relationship nodes. Mechanical relationship nodes include patient limb nodes, medical tubing nodes, hidden constraint nodes, bed boundary nodes, and equipment connection nodes. Patient limb nodes utilize the head, shoulder, wrist, hip, knee, and ankle joints continuously tracked by S110. Medical tubing nodes include nodes continuously tracked by S110 to S230, as well as nodes added at medical tubing endpoints, curvature change locations, tubing fixation points, equipment connection ends, hidden contact constraint locations, and predetermined arc length sampling locations. The predetermined arc length sampling interval is determined based on the image spatial resolution and the minimum permissible bending radius of the medical tubing.
[0050] For the section of tubing obscured by bedding, an estimated centerline is formed by connecting the medical tubing's entry and exit points, the hidden contact constraint point, and the last visible centerline before obscuration, and this centerline is marked as the estimated data.
[0051] When the projections of two medical tubings in the image intersect, the minimum distance between the corresponding center lines in the bed space coordinate system is calculated. A tubing contact node is set only when the minimum distance is not greater than the contact distance threshold determined based on the measurement error of the outer diameter and depth of the medical tubing, and there is synchronous displacement in consecutive image frames; otherwise, it is only recorded as a projection intersection.
[0052] Each mechanical relationship node records the node identifier, node type, associated object identifier, bed spatial coordinates, direction of movement, acquisition time, and image recognition confidence level, forming a spatial mechanical node set. For example, if the first infusion tube and the second monitoring line intersect in the top view image, but the depth data indicates that their heights differ by 40mm and their movements are independent, no tube contact node is established.
[0053] S320. Read the set of spatial mechanical nodes, sort them according to the arc length position of each medical pipeline node on the center line of the same medical pipeline, and determine two nodes that are adjacent along the center line and have no other mechanical relationship between them as adjacent nodes.
[0054] The arc length of the pipeline between adjacent nodes is calculated using the following formula: in, Let q represent the arc length of the pipeline between node i and node j, and q be the sorting index of the discrete points on the centerline. and Represents the 3D coordinates of discrete points on adjacent centerlines within the target pipe segment. This indicates the number of discrete points along the centerline contained in the target pipe segment.
[0055] For invisible tubing segments, the centerline length of the last fully visible segment before obstruction is used first; if no complete record exists, the length of the medical tubing registration is used after deducting the lengths of the remaining known segments, and the confidence level of the length estimate is marked.
[0056] The remaining active length of the medical tubing, which is available for patient limb movement, bedside movements, or periodic movement of medical equipment under the current constraints, is defined as the length of free tubing available for these activities, and is calculated using the following formula: Where R represents the remaining active length of the medical tubing. This represents the arc length of the pipeline between the event-affected node and the nearest limiting node. This represents the straight-line distance between two nodes. This indicates the length occupied by the hidden contact constraint, which is taken from the abnormal pipe segment length determined by S230. This indicates the length of time a nursing safety reserve is maintained, which is derived from nursing standards or nursing staff allocation.
[0057] The gripping type hidden contact constraint is configured as a local fixed node that restricts displacement in multiple directions; the pressing type hidden contact constraint is configured as a distributed clamping zone consisting of boundary nodes clamping at both ends of the abnormal pipe segment and a restricted node in the middle; the winding type hidden contact constraint is configured as a turning constraint path that includes an entry node, a bypass node, and an exit node; and the hooking type hidden contact constraint is configured as a unidirectional sliding node that allows movement in the departure direction and restricts movement in the tightening direction.
[0058] A directed node sequence that travels from the patient's limb node through hidden constraint nodes and medical tubing nodes to the tubing fixed point is defined as a displacement transmission path; a directed node sequence that travels from the bed boundary node or equipment connection node through medical tubing nodes to the patient's side tubing fixed point is defined as an external event transmission path.
[0059] When multiple candidate paths exist, all are retained, and the starting point, ending point, nodes passed through, and transmission direction of each path are recorded to form a dynamic mechanical relationship diagram. For example, when a patient grasps the first ventilator tubing with their right hand, a displacement transmission path is established between the right wrist node, the grasping node, and the endotracheal tube fixation point.
[0060] S330. A directed weighted graph consisting of mechanical relationship nodes, node connection relationships, displacement transmission direction and mechanical response parameters is defined as a dynamic mechanical relationship model. It is used to calculate the predicted displacement of each medical pipeline node and the change of the remaining active length of the medical pipeline based on the input displacement of the event to be simulated. It is not used to calculate the actual stress or contact force of the medical pipeline.
[0061] The direction of node displacement transfer is determined based on the tangential direction of the medical tubing centerline, the direction of hidden contact constraints, and the actual node motion direction in real-time perturbations; the displacement transfer ratio between adjacent nodes is calculated according to the following formula: in, This represents the proportion of displacement transferred from node i to node j. and Let represent the displacement vectors of two adjacent nodes in the effective real-time perturbation, respectively. This indicates a positive number in mm to prevent the denominator from being zero. Data is not used if node i has not reached the effective displacement threshold. In one specific embodiment, due to the influence of bedding friction and the elasticity of the pipeline itself, the displacement transfer ratio under normal unconstrained conditions... The value is between 0.3 and 0.8; when a pressure-type hidden contact constraint is identified, the displacement transfer ratio through the pressure-bearing pipe section is... It will decrease significantly to below 0.1.
[0062] The current bending state of the medical tubing is represented by the minimum bending radius corresponding to the centerline; the degree of hidden contact constraint is directly read from the S230 output value; the allowable offset range of the tubing fixing point is derived from the product parameters of the fixing device, nursing regulations, or the sum of the mean of natural fluctuations within the stable observation window and three times the standard deviation.
[0063] The initial state of the dynamic mechanical relationship model includes the current coordinates of each mechanical relationship node, the arc length of the pipeline between adjacent nodes, the remaining active length of the medical pipeline, the type and degree of hidden contact constraints, the posture of the bed, and the positions of the pipeline fixing points and equipment connection ends. The model input includes the event starting node, displacement direction, displacement amplitude, and duration. The model output includes the predicted displacement of the medical pipeline nodes, the reduction in the medical pipeline active margin, the pulling displacement of the pipeline fixing points, the predicted minimum bending radius, the tightening of hidden contact constraints, and the corresponding transmission path.
[0064] When medical tubing is rearranged, tubing fixing points are changed, the number of hidden contact constraints changes, or image recognition continuously fails for more than the allowed time, the current model is stopped and steps S210 to S330 are re-executed. The dynamic mechanical relationship model is sent to S410 for event configuration and to S420 for response simulation.
[0065] S4 specifically includes the following sub-steps: S410 reads the dynamic mechanical relationship model established by S330 and obtains the patient's recent motion records, bed-to-be-executed instructions, medical equipment operating cycle parameters, and nursing safety test parameters according to the unified system timestamp. Patient physiological activities, patient limb movements, bed actions, or medical equipment periodic movements that can generate quantifiable displacement inputs at patient limb nodes, bed boundary nodes, or equipment connection nodes are defined as events to be deduced.
[0066] Parameters for cough events are derived from image recognition of rapid chest expansion, head and neck flexion, and synchronized shoulder movements in continuous spatiotemporal imaging, or from cough event markers output by respiratory monitoring equipment. Parameters for head rotation and limb repositioning events are derived from the historical movement trajectories of the patient's limb joints. Parameters for bed posture adjustment events are derived from pending instructions or nursing plans in the bed controller. Parameters for periodic motion events of medical devices are derived from the medical device controller, or calculated from the continuous displacement trajectory of the device connection. When patient history events are unavailable, pre-configured safety test parameters by nursing staff are used, and the source of these parameters is marked in the event log.
[0067] Each event to be simulated is configured with an event identifier, event type, event start node, movement direction, movement amplitude, duration, parameter source, simulation time window, and event parameter reliability. Bed posture adjustment events use the planned execution time, and medical equipment cyclical movement events use the equipment operation cycle; for patient events where the specific occurrence time cannot be determined, the subsequent monitoring cycle is used as the simulation time window.
[0068] The event input displacement is expressed as follows: in, This represents the input displacement vector of the event to be simulated at time t. Indicates the magnitude of the event's movement. A unit vector representing the direction of movement of an event. The normalized displacement variation coefficient represents the duration of the event. The normalized displacement variation coefficient is preferably obtained by normalizing the displacement curves of similar historical events. If no historical curves are available, a piecewise linear curve representing the start, the maximum displacement, and the end is used in sequence.
[0069] For the same event, a historical baseline range and a nursing allowable upper limit range are generated separately. For example, if a patient's three most recent head rotations to the right were 18mm, 22mm, and 20mm, the median of 20mm is used as the historical baseline range, and 22mm is used as the upper limit range, forming a set of events to be extrapolated.
[0070] S420. According to a unified time step, input the event to be simulated into the corresponding event start node, and calculate the predicted displacement node by node along the displacement transmission path or external event transmission path.
[0071] For the connection from node i to node j, the predicted displacement of node j is calculated according to the following formula: in, and Let represent the predicted displacements of node i and node j at time t, respectively. This represents the proportion of displacement transferred from node i to node j. This indicates that the upstream displacement is projected based on the tangential direction of the medical pipeline centerline and the direction of the hidden contact constraint.
[0072] For locally fixed nodes, the predicted displacement in the restricted direction is set to 0. For unidirectional sliding nodes, only the displacement component in the allowed direction is retained. The turning constraint path is transmitted according to the entry and exit directions, respectively. When there are multiple transmission paths, they are calculated separately. When multiple paths converge at the same node and have the same direction, the displacement result with the larger absolute value is retained. When the directions are opposite, vector synthesis is performed, and the path that leads to the maximum dangerous response is determined as the dominant transmission path.
[0073] The reduction in the remaining active length of the medical tubing after an event relative to before the event is defined as the reduction in the active length of the medical tubing. This value represents the consumption of free tubing length, not the actual tensile force on the medical tubing. The component of the predicted displacement of the tubing fixation point in the tangential direction of the connected medical tubing is defined as the tensile displacement of the tubing fixation point. The predicted minimum bending radius of the target tubing segment after the event is taken as the tubing bending state. The reduction in the length of the free tubing on both sides of the hidden contact constraint, the active length of the bypass path, or the unidirectional sliding length is defined as the tightening of the hidden contact constraint. The above responses are calculated for each event to be simulated to form the event response prediction results.
[0074] For each target hidden contact constraint, a control model is established. The patient limb node, bed boundary node, medical pipeline centerline, inter-node arc length, pipeline fixing point, equipment connection end, nursing safety retention length, bed posture and event parameters are kept unchanged. Only the target hidden contact constraint is restored to a free connection state, and the same event is re-executed to form a control response prediction result.
[0075] When multiple hidden contact constraints exist on the same medical tubing, simulations are performed to retain all constraints, remove each constraint one by one, and remove all constraints. For example, if a patient grasps the first ventilator tubing with their right hand and the tubing is simultaneously hooked to the bed rail, the grasping constraint and the hooking constraint are removed separately to determine the contribution of the two constraints to the hazard response.
[0076] S430: Read the event response prediction results and control response prediction results, and retrieve the corresponding safety thresholds based on the medical tubing identifiers. The medical tubing activity margin safety threshold is derived from tubing layout specifications or nursing requirements; the allowable offset threshold for tubing fixation points is derived from fixation device product parameters, nursing regulations, or the upper limit of natural fluctuation of fixation points within a stable observation window; the minimum allowable bending radius of the medical tubing is derived from product parameters; and the hidden contact constraint tightening threshold is derived from the safety calibration test of the corresponding constraint state. Each safety threshold is stored in association according to medical tubing type, connection location, and fixation method, and is retrieved separately for different medical tubing types.
[0077] The state in which the same event to be simulated does not produce a threshold exceedance when the target hidden contact constraint is removed, but produces at least one threshold exceedance when the target hidden contact constraint is retained, and the exceedance response can be traced back to the target hidden contact constraint along the dominant propagation path, is defined as delayed dangerous coupling.
[0078] A threshold exceedance is determined when the reduction in the medical tubing's mobility margin reaches the corresponding upper limit, the traction displacement at the tubing's fixed point reaches the allowable offset threshold, the predicted minimum bending radius is not greater than the minimum allowable bending radius, or the tightening of the hidden contact constraint reaches the corresponding threshold. If the control response itself has already exceeded the limit, the event to be simulated is classified as its own hazard, and only the hazard increment caused by the hidden contact constraint is recorded. The hazard increment is the difference between the event response prediction result with the hidden contact constraint retained and the control response prediction result with the constraint removed; not all hazards are attributed to the hidden contact constraint.
[0079] The reliability of the simulation is jointly determined by the reliability of the event parameters, the confidence level of image recognition of key nodes in the dominant transmission path, the effectiveness of the dynamic mechanical relationship model, and the persistence of hidden contact constraints. The minimum value among the normalized results of each factor is used as a conservative value. Delayed hazardous coupling is confirmed only when the dynamic mechanical relationship model is effective, the hidden contact constraints are maintained for a predetermined number of frames, the reliability of the simulation reaches the reliability threshold, and the hazardous response exceeds the limit for more than two consecutive simulation time steps.
[0080] Each delayed hazard coupling result record includes the event identifier and parameter source to be simulated, the target hiding contact constraint identifier, the affected medical pipeline identifier, the dominant transmission path, the predicted response value, the corresponding safety threshold, the hazard increment, the predicted time to reach the threshold, the expected hazard outcome, and the simulation credibility, and is sent to S510.
[0081] For example, when the head turns 20mm to the right, the endotracheal tube movement margin in the control model decreases by 2mm but does not exceed the limit. However, when the grasping hidden contact constraint is retained, the movement margin decreases by 12mm and exceeds the limit. This confirms that there is a delayed danger coupling between the head rotation event and the grasping hidden contact constraint.
[0082] S5 specifically includes the following sub-steps: S510 reads the delay hazard coupling results output by S430, and reads medical pipeline registration information, nursing control permission table and early warning terminal online status.
[0083] The delayed hazard coupling results include the event to be simulated, target concealment contact constraints, affected medical tubing, dominant transmission path, predicted response value, corresponding safety threshold, prediction time, and simulation reliability; medical tubing registration information is entered by nursing staff when medical tubing is connected, including medical tubing type, outer diameter, registered length, connection location, fixing method, minimum allowable bending radius, and medical tubing risk attributes; the nursing control permission table is used to distinguish between non-treatment-necessary actions that are allowed to be automatically restricted and treatment actions that are prohibited from being automatically changed; the online status of the early warning terminal is returned by each terminal.
[0084] The structured data used to determine the early warning method and permissible control actions is defined as the risk handling parameter set. For each delayed hazard coupling, the risk assessment value is calculated using the following formula: in, Indicates the risk assessment value. Indicates the degree to which the threshold is approached or exceeded. Indicates the urgency and timeliness of the danger. This indicates the risk attribute value of medical tubing. This indicates the reliability of the inference output by S430. to This represents the weights that sum to 1. , , and All normalized to 0 to 1; Determined based on the ratio of the predicted response value to the safety threshold. The value is determined based on the ratio of the predicted time to reach the threshold to the nursing intervention time. Based on a preset level mapping corresponding to the risk attributes of medical tubing, in a preferred embodiment, the four weights are respectively set as follows: =0.35、 =0.35、 =0.2、 =0.1; when the calculated risk assessment value When, it was determined to be an emergency level risk; when When, it is determined to be an intervention-level risk; when At that time, it was identified as a warning level risk.
[0085] If the predicted response enters the vicinity of the safety threshold but has not yet exceeded the limit, or if the confidence level of the inference is insufficient to support the automatic restriction action, it is identified as a warning level risk; if the predicted response exceeds the limit and the prediction time is longer than the nursing intervention time, it is identified as an intervention level risk; if the predicted response exceeds the limit and the prediction time is not longer than the nursing intervention time, or if the affected medical tubing is a high-risk medical tubing, it is identified as an emergency level risk.
[0086] Nursing intervention time is set by nursing staff based on the type of medical tubing and the ward's response distance. When multiple outcomes exist for the same medical tubing, the highest risk level is used. When multiple medical tubings are affected simultaneously, the outcome with the highest risk attribute and shortest prediction time is prioritized.
[0087] The risk management parameter set includes risk record identifiers, risk levels, hidden contact restraint type and location, affected medical tubing identifiers, predicted hazardous outcomes, predicted timeframes, permissible control actions, prohibited actions, and warning receiving terminals. For example, if a grasping-type hidden contact restraint is expected to cause excessive traction displacement of the endotracheal tube fixation point after 8 seconds, it is classified as an emergency risk.
[0088] S520: Read the risk management parameter set and generate graded early warning instructions according to risk level and hidden contact constraint type. For alert-level risks, send inspection prompts to the bedside display terminal and nursing terminal, displaying the affected medical tubing, the location of hidden contact constraints, and the event to be simulated. For intervention-level risks, further generate nursing management instructions, requiring release of grips, removal of pressure, untangling of entanglements, or disengagement, and suspending non-treatment-necessary bedside actions corresponding to the event to be simulated. For emergency-level risks, activate bedside audible and visual alarms, nursing terminal pop-ups, and remote monitoring terminal alarms, displaying the expected dangerous outcome and predicted remaining time.
[0089] Record the tiered early warning instruction identifier, sending time, receiving terminal, terminal confirmation time, and medical staff confirmation time. If no confirmation is received within the preset confirmation time, the original risk level remains unchanged, but the scope of the tiered early warning instruction is expanded or the intensity of the warning output is increased; warnings corresponding to the alert level are forwarded to the backup nursing terminal, warnings corresponding to the intervention level are forwarded to the remote monitoring terminal, and bedside audible and visual alarms corresponding to the emergency level continue to be output. The preset confirmation time is configured by nursing staff according to the risk level and the ward's communication conditions.
[0090] Based on the nursing control permission table, device control instructions are generated, restricting only non-treatment-necessary bed movements or medical device-assisted movements that are coupled with delayed risks and have been pre-authorized. This includes pausing movements that have not yet begun, such as turning over, back lifting, or legboard adjustment; preventing automatic bed posture switching in non-emergency nursing modes; or pausing device-connected assisted movements that do not affect basic treatment output. Before execution, the system verifies whether the target movement is on the allowed automatic restriction list and confirms that there are no emergency treatment instructions, manually enforced instructions, or prohibited automatic control states. If the verification fails, only a tiered warning instruction is output.
[0091] Equipment control commands must not stop respiratory support, intravenous infusion, monitoring, or drainage, nor alter essential life-sustaining treatment parameters. After execution, the execution status returned by the controller should be read; if a success status is not returned, a control failure message should be appended to the tiered warning command. For example, a bed-lifting maneuver that has not yet been executed and would tighten the first ventilator tubing can be paused, but the ventilator's ventilation maintenance actions must not be stopped.
[0092] S530. After medical staff complete the treatment, the nursing terminal inputs the treatment completion information, or the continuous spatiotemporal images confirm that the patient's hand has left the target medical tubing, the compressed limb has been removed, the entanglement path has been unwound, or the hooked position has been released. This serves as the condition for initiating the risk release verification. The information is not directly used as the risk release conclusion.
[0093] Before performing the predetermined micro-displacement again, confirm that the patient's posture, the target area of the bed, the medical tubing markings, the tubing fixing points, and the bedding covering method still meet the applicable conditions of the target baseline micro-perturbation response template, and confirm that the patient has not made any active movements, the medical staff have left the area affected by the perturbation, the bed controller is operating normally, and there is no signal prohibiting bed movement.
[0094] Perform a predetermined micro-displacement according to the same target area, direction, amplitude, speed, and duration as the target baseline perturbation response template to obtain the release verification response. The release verification response includes the propagation delay of bedding deformation, displacement of medical tubing nodes, node movement sequence, offset of tubing fixation points, recovery time, and recovery path.
[0095] The standardized variance of each key response feature was calculated according to the method of S220, and the devalidation residuals were calculated according to the following formula: in, This indicates the removal of the verification residual. Represents the set of key response features. This represents the standardized variance of the j-th key response feature.
[0096] Once all key response characteristics return to the allowable fluctuation range of the target baseline perturbation response template, the verification response input S210 will be deactivated, and S210 to S430 will be re-executed. Risk clearance is confirmed only when S230 no longer identifies the corresponding hidden contact constraint and S430 no longer outputs the corresponding delayed hazardous coupling result. Based on the risk record identifier, graded early warning instruction identifier, and equipment control instruction identifier output by S520, the corresponding instructions are revoked.
[0097] After the risk is eliminated, if at least three valid elimination verification responses are obtained consecutively, and the patient's posture, medical tubing layout, tubing fixation points, bedding coverage method, and target area of the bed remain unchanged, the valid elimination verification responses are incorporated into the original valid baseline perturbation response sample set. The mean, standard deviation, and allowable fluctuation range of each response characteristic are recalculated according to method S130 to form an updated baseline perturbation response template. If the patient's posture, medical tubing fixation method, or bedding coverage method changes, S110 to S130 are repeated to establish a new template.
[0098] Based on the latest continuous spatiotemporal images, update the patient's limb nodes, medical tubing centerline, medical tubing nodes, tubing fixation points, equipment connection ends, and remaining active length of the medical tubing. Delete the hidden constraint nodes and connection relationships that have been released, update the dynamic mechanical relationship model, and then return to S210 to continue monitoring.
[0099] For example, if the third medical pipeline node resumes movement after the grip is released, but the fifth medical pipeline node still experiences displacement propagation interruption, it indicates that the hidden contact constraint has not completely disappeared. Therefore, the warning should not be released, and this response should not be written into the baseline perturbation response template.
[0100] Example 2: Figure 2 As shown, this embodiment provides a system for monitoring abnormal behavior in critically ill patients based on spatiotemporal images, including: The spatiotemporal imaging and benchmark establishment module is used to acquire continuous spatiotemporal images of the patient, bedding, hospital bed and medical tubing, identify the joints of the patient's limbs, nodes of the medical tubing and tubing fixation points, determine the target area of the hospital bed, control the target area of the hospital bed to perform predetermined micro-displacements, and establish a benchmark micro-perturbation response template. The hidden contact constraint identification module is used to match the target reference perturbation response template, obtain the real-time perturbation response sequence, calculate the response differences between bedding and medical tubing, identify gripping, pressing, wrapping or hooking hidden contact constraints, and form a set of hidden contact constraints. The dynamic mechanical relationship modeling module is used to establish a spatial mechanical node set based on the hidden contact constraint set, patient limb joints, medical tubing nodes, tubing fixing points, bed boundaries and equipment connection ends, determine the remaining active length and transmission path of the medical tubing, and construct a dynamic mechanical relationship model. The delayed hazard simulation module is used to generate a set of events to be simulated based on patient motion records, bedside instructions to be executed, and medical equipment operating parameters. The set of events to be simulated is then input into a dynamic mechanical relationship model to obtain event response prediction results and control response prediction results, and to determine the delayed hazard coupling results. The risk handling and de-coupling verification module is used to determine the risk level based on the delayed hazard coupling result, generate graded early warning instructions and pre-authorized equipment control instructions, and control the target area of the bed to perform a predetermined micro-displacement again after handling. After confirming that the corresponding hidden contact constraint and delayed hazard coupling have disappeared, the corresponding instructions are revoked, and the baseline perturbation response template and dynamic mechanical relationship model are updated.
[0101] All the above formulas are performed using dimensionless numerical calculations; the relevant formulas are based on empirical models that approximate the real situation, obtained through extensive data collection and software simulation fitting. The preset parameters and thresholds involved in the formulas can be conventionally set and adjusted by those skilled in the art according to the physical constraints of the actual application scenario.
[0102] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0103] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0104] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for monitoring abnormal behavior in critically ill patients based on spatiotemporal imagery, characterized in that, Includes the following steps: S1. Acquire continuous spatiotemporal images of the patient, bedding, bed, and medical tubing; identify joints of the patient's limbs and nodes of the medical tubing; determine the target area of the bed; perform predetermined micro-displacement; and establish a baseline perturbation response template. S2. Match the target reference perturbation response template, perform a predetermined micro-displacement, obtain the real-time perturbation response sequence and calculate the response difference, identify gripping, pressing, entanglement or hooking type hidden contact constraints, and form a set of hidden contact constraints. S3. Based on the hidden contact constraint set, the patient's limb joints, the medical tubing nodes and their fixed points and equipment connection ends, establish a spatial mechanical node set, determine the node connections, the remaining active length of the medical tubing and the transmission path, and construct a dynamic mechanical relationship model. S4. Generate a set of events to be simulated based on the patient's motion records, bedside instructions to be executed, and medical equipment operating parameters. Input the dynamic mechanical relationship model to obtain the event response prediction results and the control response prediction results, and determine the delayed danger coupling results.
2. The method for monitoring abnormal behavior of critically ill patients based on spatiotemporal imagery according to claim 1, characterized in that, Also includes: S5. Determine the risk level based on the delayed hazard coupling result, and generate graded early warning instructions and pre-authorized equipment control instructions; After handling, the predetermined micro-displacement is executed again. Once it is confirmed that the hidden contact constraint has disappeared and there is no delayed dangerous coupling, the corresponding command is revoked and the benchmark perturbation response template and dynamic mechanical relationship model are updated.
3. The method for monitoring abnormal behavior in critically ill patients based on spatiotemporal imagery according to claim 1, characterized in that, S1 specifically includes: Collect continuous spatiotemporal images of the patient, bedding, bed, and medical tubing; establish a spatial coordinate system for the bed; identify the joints of the patient's limbs, nodes of the medical tubing, and fixed points of the tubing; generate a baseline object tracking sequence; and determine the target area of the bed. When the patient has no active movement and the bed drive is normal, the target area of the bed is controlled to perform a predetermined micro-displacement, and the responses of the bedding, medical tubing nodes and tubing fixed points are recorded to form a baseline micro-perturbation response sample sequence. The baseline perturbation response sample sequence was time-aligned and displacement-normalized to extract the propagation delay of bedding deformation, node motion sequence, recovery time and recovery path. A baseline perturbation response template was established according to the patient's posture, the target area of the bed and the medical pipeline markings.
4. The method for monitoring abnormal behavior of critically ill patients based on spatiotemporal imagery according to claim 1, characterized in that, S2 specifically includes: Based on the patient's posture, the target area of the bed, the medical tubing markings, and the bedding covering method, a target baseline perturbation response template is matched, and it is controlled to perform a predetermined micro-displacement to obtain a real-time perturbation response sequence. Align and normalize the real-time perturbation response sequence with the target baseline perturbation response template, calculate the difference in bedding deformation propagation delay, the difference in medical pipeline node displacement, the difference in node motion order, the difference in pipeline fixed point offset, and the deviation of the recovery path to form a real-time difference response feature set. Abnormal pipe segments are determined based on the real-time differential response feature set. By combining node displacement, movement sequence, curvature change and patient limb position, grasping, pressing, entanglement or hooking hidden contact constraints are identified to form a hidden contact constraint set.
5. The method for monitoring abnormal behavior of critically ill patients based on spatiotemporal imagery according to claim 1, characterized in that, S3 specifically includes: Read the hidden contact constraint set, combine it with continuous spatiotemporal images, bed control data and medical pipeline registration information, and set patient limb nodes, medical pipeline nodes, hidden constraint nodes, bed boundary nodes and equipment connection nodes to form a spatial mechanical node set; Determine adjacent nodes according to the centerline of the medical pipeline, calculate the arc length and remaining active length, configure hidden contact constraints as local fixed nodes, clamping areas, turning constraint paths or unidirectional sliding nodes, establish displacement transmission paths and external event transmission paths, and form a dynamic mechanical relationship diagram.
6. The method for monitoring abnormal behavior of critically ill patients based on spatiotemporal imagery according to claim 5, characterized in that, Also includes: Based on the dynamic mechanical relationship diagram and real-time perturbation node displacement, the displacement transmission direction, displacement transmission ratio, and allowable offset range of fixed points are configured to establish a dynamic mechanical relationship model.
7. The method for monitoring abnormal behavior in critically ill patients based on spatiotemporal imagery according to claim 1, characterized in that, S4 specifically includes: Read the dynamic mechanical relationship model, and based on the patient's motion records, the instructions to be executed by the bed, and the operating parameters of the medical equipment, configure the event start node, direction, amplitude, duration, and simulation time window to generate a set of events to be simulated; The event set to be simulated is input into the dynamic mechanical relationship model. The predicted displacement of pipeline nodes, the reduction of movable margin, the pulling displacement of fixed points, and the tightening of hidden contact constraints are calculated along the displacement transmission path or the external event transmission path to form the event response prediction results. The hidden contact constraints are removed to obtain the control response prediction results.
8. The method for monitoring abnormal behavior of critically ill patients based on spatiotemporal imagery according to claim 7, characterized in that, Also includes: By comparing the event response prediction results with the control response prediction results, if a threshold exceeds the limit while the hidden contact constraint is retained and the control state does not exceed the limit, and the exceeding response can be traced back to the hidden contact constraint, a delayed hazard coupling is determined, and a delayed hazard coupling result is formed.
9. The method for monitoring abnormal behavior of critically ill patients based on spatiotemporal imagery according to claim 2, characterized in that, S5 specifically includes: Read the delayed hazard coupling results, pipeline registration information and control permissions, combine the threshold proximity, hazard time, pipeline risk attributes and simulation credibility to determine the risk level, and generate a risk treatment parameter set; Based on the risk management parameter set, a graded early warning instruction is generated, and the list of permitted automatic restrictions is verified. Only the corresponding non-treatment-necessary bed actions or medical equipment-assisted actions are suspended, without changing the basic life-sustaining treatment parameters. After handling, the predetermined micro-displacement is executed again according to the target benchmark perturbation response template to obtain the release verification response. Hidden contact constraints are re-identified and delayed dangerous coupling is deduced. After the corresponding constraints and coupling results disappear, the command is revoked and the benchmark perturbation response template and dynamic mechanical relationship model are updated.
10. A system for monitoring abnormal behavior in critically ill patients based on spatiotemporal images, employing the method for monitoring abnormal behavior in critically ill patients based on spatiotemporal images as described in any one of claims 1 to 9, characterized in that, include: The spatiotemporal image and benchmark establishment module is used to acquire continuous spatiotemporal images of patients, bedding, hospital beds and medical tubing, and establish benchmark perturbation response templates; The hidden contact constraint identification module is used to match the target reference perturbation response template, obtain the real-time perturbation response sequence, and form a hidden contact constraint set. The dynamic mechanical relationship modeling module is used to establish a set of spatial mechanical nodes based on the hidden contact constraint set, patient limb joints, medical tubing nodes, tubing fixing points, bed boundaries and equipment connection ends, and to construct a dynamic mechanical relationship model. The delayed danger simulation module is used to generate a set of events to be simulated based on the patient's motion records, the instructions to be executed on the hospital bed, and the operating parameters of the medical equipment, and input the set of events to be simulated into the dynamic mechanical relationship model; The risk handling and de-escalation verification module is used to determine the risk level based on the delayed hazard coupling result, and generate graded early warning instructions and pre-authorized equipment control instructions.
