Method and system for collecting and remotely monitoring pressure data of an abdominal belt for external version of breech presentation

Abstract

This invention provides a method and system for acquiring and remotely monitoring abdominal pressure data during breech external valgus reversal, relating to the field of medical monitoring technology. The method includes acquiring a pressure event stream with temporal and locational information through a pressure sensing layer within the abdominal binder, constructing a pressure propagation path map, and calculating path offset features. Multi-dimensional features are extracted from the pressure event stream to generate an operational behavior feature vector, and a state transition model is constructed. The model extrapolates and predicts subsequent operations, and a risk assessment result is generated and uploaded to a remote platform based on a safety threshold. The platform then generates and issues operational intervention commands, achieving real-time, quantitative monitoring and remote safety early warning of the breech external valgus reversal operation process.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, and in particular to a method and system for collecting and remotely monitoring pressure data of external abdominal binders in breech presentation. Background Technology

[0002] In obstetric clinical practice, external cephalic version (ECV) is an important procedure used to correct breech presentation to cephalic presentation. The success and safety of this procedure highly depend on the surgeon's experience and technique. However, current techniques lack objective and quantitative data support for the procedure. The surgeon relies entirely on personal experience to perceive the magnitude, direction, and changes in pressure applied, and assesses the fetal position and response through palpation. There is no real-time, continuous recording of pressure data throughout the entire process; the pressure distribution, application trajectory, and temporal dynamics cannot be accurately captured and retrospectively analyzed. This results in significant individual variability and subjectivity in the procedure, making standardized assessment and teaching difficult. Furthermore, due to the lack of real-time monitoring and early warning mechanisms, the control of operational risks depends entirely on the surgeon's immediate judgment. It is difficult to detect and intervene promptly when improper pressure is applied or when the fetus exhibits a stress response.

[0003] This conventional approach has significant drawbacks. On the one hand, the reliance on experience-based procedures makes technical training difficult, resulting in a long learning curve for novice physicians and inconsistent skill levels, which hinders the standardized promotion and application of the technique. On the other hand, the lack of objective data feedback and remote monitoring means that potential risks (such as excessive local pressure or continuous deviation in the direction of force) cannot be identified and warned of in real time during the procedure, potentially increasing the risk of complications such as discomfort in pregnant women or fetal distress, thus limiting the safe application of this technique in a wider range of medical scenarios. Summary of the Invention

[0004] This invention provides a method and system for acquiring and remotely monitoring abdominal pressure data in breech external valgus positions, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for acquiring and remotely monitoring abdominal binder pressure data in cases of breech presentation with external valgus support, comprising: By embedding a pressure sensing layer in the abdominal binder, real-time pressure signals and timestamps of each sensing unit during the external breech inversion operation are collected, forming a pressure event stream with time sequence annotation and location identifier; Based on the pressure transmission timing and amplitude differences between adjacent sensing units, a pressure propagation path map representing the direction of force application is constructed, and the path offset characteristics are calculated. Based on the pressure event stream, pressure peak, pressure change slope and sensor unit activation sequence are extracted using an adaptive time window. Combined with the path offset features, an operation behavior feature vector is generated. An operation behavior state transition model is constructed based on the operation behavior feature vector sequence of continuous time windows. The subsequent feature vectors are extrapolated and predicted based on the state transition model. A risk assessment result is generated by combining the prediction confidence and the pressure safety limit. The risk assessment result and the pressure event stream are synchronously uploaded to the remote monitoring platform. The remote monitoring platform generates an operation intervention command based on the risk assessment result and sends it to the operation terminal.

[0006] By embedding a pressure-sensing layer within the abdominal binder, real-time pressure signals and timestamps from each sensing unit during external breech inversion are collected, forming a pressure event stream with time-series and location markers, including: Based on the abdominal anatomical structure, the sensing units in the pressure sensing layer are divided into zones and classified, and each sensing unit is assigned a unique location identifier corresponding to its anatomical location. The average pressure of each sensing unit at rest before force application is used as the individualized baseline. During the acquisition process, the baseline difference operation is continuously performed on the real-time pressure signal to extract the effective force application increment signal after removing the static pre-pressure. The event trigger boundary is determined based on the amplitude change characteristics of the effective force increment signal. The effective acquisition window is dynamically captured with the event trigger boundary as the start and end time. Within the effective acquisition window, the timestamps of all sensing units are aligned with a unified time base to eliminate multi-channel acquisition timing deviation. The effective force increment signal amplitude of each sensing unit within the effective acquisition window is combined with the alignment timestamp, so that each event entry carries a location identifier, effective force increment amplitude and alignment timestamp simultaneously, forming a pressure event stream with time sequence labeling and location labeling.

[0007] Based on the differences in pressure transmission timing and amplitude between adjacent sensing units, a pressure propagation path map representing the direction of force application is constructed, and path offset characteristics are calculated, including: Based on the pressure event flow, the first activation time of the pressure signal of each sensing unit is extracted. The direction of the directed edge propagation is determined by the positive and negative directions of the difference between the first activation times of adjacent sensing units. The sum of the absolute values ​​of the difference between the first activation times and the absolute values ​​of the peak amplitude is used as the propagation cost of the directed edge. The directed edges with each sensing unit as a node and the propagation cost and propagation direction as attributes are connected to adjacent nodes to form a pressure propagation path graph. Based on the pressure propagation path diagram, starting from the force application node, nodes are gradually expanded from small to large while maintaining path connectivity. The path with the minimum cumulative propagation cost to the force termination node is extracted as the main pressure propagation path. The corresponding spatial coordinates of each node on the main path are retrieved by the position identifiers of each node, arranged in the order of cumulative propagation cost, and then fitted to obtain the main path axis. The distance between the starting endpoint coordinates of the main path axis and the pre-calibrated standard force application starting coordinates is used as the starting offset of the path offset feature. The angle between the overall direction vector of the main path axis and the pre-calibrated standard force application direction vector is used as the directional offset of the path offset feature. The starting offset and the directional offset are combined to obtain the path offset feature.

[0008] Based on the pressure event stream, pressure peaks, pressure change slopes, and sensor unit activation sequences are extracted using an adaptive time window. These are then combined with the path offset features to generate operational behavior feature vectors. An operational behavior state transition model is constructed based on the operational behavior feature vector sequence within a continuous time window, including: A covariance matrix is ​​constructed for the pressure event stream and principal component analysis is performed. The time series of the first principal component is extracted and the envelope is calculated. The adaptive time window is defined by the time when the envelope enters the peak interval and the time when it leaves the peak interval. Within each adaptive time window, spatial interpolation is performed on the pressure peak of each sensing unit to generate a pressure distribution surface. The maximum modulus of the gradient field of the pressure distribution surface is calculated as the pressure change slope. Sensing units whose pressure peak exceeds the middle value of the window are identified and sorted by time to form a high-pressure activation sequence. The minimum convex hull is calculated based on the spatial coordinates of the sensing unit corresponding to the high-pressure activation sequence. The distance between the geometric center of the minimum convex hull and the predetermined standard center is used as the center offset, and the ratio of the area of ​​the minimum convex hull to the predetermined standard area is used as the coverage offset. This is then concatenated with the pressure change slope and the high-pressure activation sequence, and combined with the path offset features to generate an operation behavior feature vector. The operation behavior feature vectors of continuous time windows are used to construct a time series matrix. The Pearson correlation coefficient between feature vectors within the sliding window is calculated and the state boundaries are divided. The state representative vector of each state segment is calculated. The number of transitions and the transition time between the state representative vectors are statistically analyzed to construct the operation behavior state transition model.

[0009] The operation behavior feature vectors within a continuous time window are used to construct a time series matrix. The Pearson correlation coefficient between feature vectors within the sliding window is calculated, and state boundaries are defined. The state representative vector for each state segment is calculated, and the number of transitions and transition times between state representative vectors are statistically analyzed to construct an operation behavior state transition model, including: The operation behavior feature vectors of continuous time windows are arranged in chronological order to form a time series matrix. A sliding window scan is performed on the time series matrix to calculate the Pearson correlation coefficient and construct a correlation coefficient matrix. The trace of the correlation coefficient matrix is ​​calculated to obtain the trace value time series. Change point detection is performed on the trace value time series, and the detected change point position is determined as the state boundary. The time series matrix is ​​segmented by the state boundary to obtain multiple state segments. The centroid of the operation behavior feature vector of each state segment is calculated as the state representative vector. The time span of each state segment is calculated as the state dwell time. Construct a state transition sequence, extract representative vectors of adjacent states in time sequence to form transition tuples, count the frequency of occurrence of the same transition tuple as the transition count, calculate the time span between the preceding state segment and the subsequent state segment corresponding to the same transition tuple and calculate the average value as the average transition time. State nodes are established using state representation vectors, with the state dwell time as a node attribute. Directed transition edges are established using transition tuples. The normalized transition rate is obtained by dividing the transition count by the state dwell time of the preceding state segment and used as the edge weight. The average transition time is used as the edge attribute. An operation behavior state transition model is constructed based on the state nodes and the directed transition edges.

[0010] Based on the state transition model, subsequent feature vectors are extrapolated and predicted. A risk assessment result is generated by combining the prediction confidence level and the stress safety upper limit. This risk assessment result, along with the stress event stream, is then synchronously uploaded to the remote monitoring platform, including: Extract the current state node from the state transition model, trace back the historical transition path to identify the cyclic transition sub-path, and determine whether the current state node is located within the cyclic transition sub-path. If it is located within the cyclic, extract the state representative vectors of each state node in the cyclic to form a cyclic feature sequence, and periodically extrapolate the cyclic feature sequence to obtain the predicted feature vector. If it is not located within the cyclic, extract the state representative vector of the successor state node with the largest normalized transition rate as the predicted feature vector. For cyclic transfer prediction, the ratio of the number of repeated cycles to the total number of historical transfers is used as the prediction confidence level; for non-cyclic transfer prediction, the difference between the maximum and the second largest normalized transfer rate is used as the prediction confidence level. The pressure component is extracted from the predicted feature vector, the difference between the pressure component and the pressure safety upper limit is calculated to obtain the pressure over-limit value, the pressure over-limit value is multiplied by the prediction confidence level to obtain the confidence weighted over-limit value, and the confidence weighted over-limit value is mapped to a preset risk partition to obtain the risk assessment result. The upload time interval is determined based on the risk assessment results. The stress event stream is sampled according to the upload time interval. The risk assessment results are appended to the sampled stress event stream and uploaded to the remote monitoring platform.

[0011] A second aspect of the present invention provides a system for acquiring and remotely monitoring abdominal binder pressure data in cases of breech presentation with external valgus, comprising: The pressure acquisition unit is used to acquire real-time pressure signals and timestamps from each sensing unit during the external breech reversal operation by means of a pressure sensing layer embedded in the abdominal binder, forming a pressure event stream with time sequence annotations and location identifiers; The path construction unit is used to construct a pressure propagation path map representing the direction of force application based on the pressure transmission timing and amplitude differences between adjacent sensing units, and to calculate the path offset characteristics. The feature extraction unit is used to extract pressure peaks, pressure change slopes and sensing unit activation sequences based on the pressure event stream using an adaptive time window, generate operation behavior feature vectors by combining the path offset features, and construct an operation behavior state transition model based on the operation behavior feature vector sequence of continuous time windows. The risk assessment unit is used to extrapolate and predict subsequent feature vectors based on the state transition model, generate risk assessment results by combining the prediction confidence and the pressure safety upper limit, and synchronously upload the risk assessment results and the pressure event stream to the remote monitoring platform. The instruction generation unit is used by the remote monitoring platform to generate operation intervention instructions based on the risk assessment results and send them to the operation terminal.

[0012] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0014] This method enables high-precision, multi-dimensional, real-time data acquisition of abdominal binder pressure during external cephalic version. By embedding a pressure sensing layer of the abdominal binder, a pressure event stream with precise temporal and spatial location information can be acquired synchronously, providing a complete and reliable data foundation for subsequent analysis.

[0015] Based on the collected pressure event stream, a pressure propagation path map representing the direction of force application can be dynamically constructed and its offset characteristics can be calculated. This technology can effectively identify the direction of force application, force diffusion mode, and potential force line offset, providing key evidence for judging whether the operation technique is correct and whether the force is concentrated. By combining the pressure peak value, rate of change, and sensor activation sequence extracted by the adaptive time window, a feature vector that comprehensively describes the behavior of a single operation can be generated.

[0016] By constructing a state transition model for operational behavior using feature vector sequences from continuous time windows, risk assessment results can be generated in real time based on the prediction results of the state transition model, combined with prediction confidence and preset safety stress thresholds. This risk assessment, along with the raw stress data, is synchronously uploaded to a remote monitoring platform, achieving deep collaboration between local operation and remote monitoring. Remote experts can make decisions based on quantified risk levels and detailed data streams, generating precise operational intervention instructions and issuing them to the operational terminals. This provides remote guidance and safety warnings at critical moments, effectively improving the safety and success rate of external breech inversion procedures. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the method for collecting and remotely monitoring abdominal binder pressure data in breech position according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating the method for calculating path offset features according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0020] Figure 1 This is a flowchart illustrating the method for acquiring and remotely monitoring abdominal binder pressure data in breech position according to an embodiment of the present invention. Figure 1 As shown, the method includes: By embedding a pressure sensing layer in the abdominal binder, real-time pressure signals and timestamps of each sensing unit during the external breech inversion operation are collected, forming a pressure event stream with time sequence annotation and location identifier; Based on the pressure transmission timing and amplitude differences between adjacent sensing units, a pressure propagation path map representing the direction of force application is constructed, and the path offset characteristics are calculated. Based on the pressure event stream, pressure peak, pressure change slope and sensor unit activation sequence are extracted using an adaptive time window. Combined with the path offset features, an operation behavior feature vector is generated. An operation behavior state transition model is constructed based on the operation behavior feature vector sequence of continuous time windows. The subsequent feature vectors are extrapolated and predicted based on the state transition model. A risk assessment result is generated by combining the prediction confidence and the pressure safety limit. The risk assessment result and the pressure event stream are synchronously uploaded to the remote monitoring platform. The remote monitoring platform generates an operation intervention command based on the risk assessment result and sends it to the operation terminal.

[0021] By embedding a pressure-sensing layer within the abdominal binder, real-time pressure signals and timestamps from each sensing unit during external breech inversion are collected, forming a pressure event stream with time-series and location markers, including: Based on the abdominal anatomical structure, the sensory units in the pressure sensing layer are divided into zones and classified, and each sensory unit is assigned a unique location identifier corresponding to its anatomical location. The average pressure of each sensing unit at rest before force application is used as the individualized baseline. During the acquisition process, the baseline difference operation is continuously performed on the real-time pressure signal to extract the effective force increment signal after removing the static pre-pressure. The event trigger boundary is determined based on the amplitude change characteristics of the effective force increment signal. The effective acquisition window is dynamically captured with the event trigger boundary as the start and end time. Within the effective acquisition window, the timestamps of all sensing units are aligned with a unified time base to eliminate multi-channel acquisition timing deviation. The effective force increment signal amplitude of each sensing unit within the effective acquisition window is combined with the alignment timestamp, so that each event entry carries a location identifier, effective force increment amplitude and alignment timestamp simultaneously, forming a pressure event stream with time sequence labeling and location labeling.

[0022] During the data acquisition phase of the external breech reversal procedure, the pressure sensing layer is composed of a flexible piezoresistive film array. This array contains 128 sensing units, which are embedded in the abdominal binder lining in an 8×16 matrix layout. Each sensing unit is made of a pressure-sensitive polymer material, and the resistance value changes linearly with pressure, with a response time of less than 5 milliseconds.

[0023] The zoning and classification are based on the anatomical features of the pregnant woman's abdomen. The abdomen is divided into five functional regions: the fundal region, the periumbilical region, the suprapubic region, and the bilateral iliac fossa regions. The fundal region corresponds to sensing units numbered A01 to A24, the periumbilical region to B01 to B32, the suprapubic region to C01 to C20, and the left and right iliac fossa regions to D01 to D26 and E01 to E26, respectively. The location identifier of each sensing unit consists of a region code and a serial number; for example, B15 represents unit number 15 in the periumbilical region. This identifier is fixed in the address mapping table of the hardware controller during the manufacturing stage.

[0024] During the resting phase before operation, pressure data is collected for 30 consecutive seconds. The average pressure during this period is calculated for each sensing unit as the individualized baseline value for that unit. Due to differences in abdominal circumference, subcutaneous fat thickness, and abdominal binder binding among different pregnant women, the baseline values ​​of each unit fluctuate between 200 Pa and 800 Pa. During real-time acquisition, the hardware circuit obtains the raw pressure value at a sampling frequency of 1000 Hz. The controller immediately performs differential calculation, that is, subtracts the baseline value of the corresponding unit from the current pressure value to obtain the effective force increment signal. When the increment signal is less than 30 Pa, it is judged as noise and set to zero, and only the increment generated by the actual force is retained.

[0025] The event trigger boundary is determined based on the amplitude change rate within a sliding time window. A sliding window of 100 milliseconds is set, and the standard deviation of the incremental signal within the window is calculated. When the standard deviation of any sensing unit suddenly exceeds three times the standard deviation during the resting period, it is determined to be the event trigger start boundary. After triggering, all units are continuously monitored. When the incremental signal of all units is below 100 Pa for 500 milliseconds consecutively, it is determined to be the event end boundary. The effective acquisition window is the time period from 200 milliseconds before the start boundary to 200 milliseconds after the end boundary. This design ensures the capture of the complete force application process.

[0026] Due to timing deviations in the hardware scanning mechanism, multi-channel acquisition uses a row-column scanning method to read the array. A complete scan cycle takes 8 milliseconds. To eliminate this deviation, a unified time base alignment is performed on all timestamps within the effective acquisition window. Specifically, the first trigger time of cell A01 is used as the zero-time reference point, and the timestamps of other cells are subtracted from their respective scan delay times. The scan delay is determined by the row and column position of the cell in the matrix. For example, the delay of the cell in the m-th row and n-th column is... With millisecond alignment, all timestamps share the same time origin, reducing timing errors to the sub-millisecond level.

[0027] The data structure of the pressure event stream is stored in the form of a triple array. Each event entry includes a location identifier field, an effective force increment amplitude field, and an alignment timestamp field. For example, an entry record of {B15, 1250Pa, 1.234s} indicates that a force increment of 1250 Pa was detected in cell 15 of the periumbilical region 1.234 seconds after the start of the event. The event stream is arranged in ascending order of timestamps, forming a complete spatiotemporal pressure evolution sequence, providing a structured data foundation for subsequent path analysis and feature extraction.

[0028] Figure 2 This is a flowchart illustrating the method for calculating path offset features according to an embodiment of the present invention. Based on the differences in pressure transmission timing and amplitude between adjacent sensing units, a pressure propagation path map representing the direction of force application is constructed, and path offset features are calculated, including: According to the pressure event flow, the first activation time of the pressure signal of each sensing unit is extracted. The positive and negative directions of the difference between the first activation times of adjacent sensing units are used to determine the direction of the directed edge propagation. The sum of the absolute values ​​of the difference between the first activation times and the absolute values ​​of the peak amplitude is used as the propagation cost of the directed edge. Each sensing unit is used as a node, and directed edges with propagation cost and propagation direction as attributes are connected to adjacent nodes to form a pressure propagation path graph; Based on the pressure propagation path diagram, starting from the force application node, nodes are gradually expanded from small to large while maintaining path connectivity. The path with the minimum cumulative propagation cost to the force termination node is extracted as the main pressure propagation path. The corresponding spatial coordinates of each node on the main path are retrieved by the position identifiers of each node, arranged in the order of cumulative propagation cost, and then fitted to obtain the main path axis. The distance between the starting endpoint coordinates of the main path axis and the pre-calibrated standard force application starting coordinates is used as the starting offset of the path offset feature. The angle between the overall direction vector of the main path axis and the pre-calibrated standard force application direction vector is used as the directional offset of the path offset feature. The starting offset and the directional offset are combined to obtain the path offset feature.

[0029] During external breech reversal, the pressure-sensing layer embedded in the abdominal binder contains an array of capacitive pressure sensors encapsulated in flexible material. Each sensing unit is distributed across the abdominal contact area at a grid spacing of 5mm x 5mm. Each sensing unit generates a capacitance change when pressure is applied, which is converted into a digital value of 0-4095 by a signal conditioning circuit, with a sampling frequency set to 100Hz. When the operator applies external force, each sensing unit is activated sequentially and outputs a pressure signal, with the timestamp recording accuracy reaching the 10ms level.

[0030] The initial activation time is determined using a dynamic threshold method. After filtering the pressure signals of each sensing unit using a 5-point moving average, when the pressure values ​​of three consecutive sampling points exceed the baseline noise mean plus twice the standard deviation, the timestamp of the first sampling point exceeding the threshold is taken as the initial activation time of that unit. Using the sensing unit number (i, j) as the node identifier, the activation time difference Δt between adjacent units (i, j) and (i+1, j) is calculated. If Δt > 0, it indicates that the pressure is transmitted from (i, j) to (i+1, j), and the direction of the directed edge is determined accordingly; if Δt < 0, the direction is reversed.

[0031] The quantification of propagation cost comprehensively considers time delay and pressure gradient. Let P1 be the peak pressure of cell (i, j) and P2 be the peak pressure of cell (i+1, j). Then, the propagation cost of this directed edge is defined as |Δt| + |P1 - P2|. The time delay reflects the propagation speed of the pressure wave, and the peak difference reflects the degree of energy attenuation. The sum of the two can effectively characterize the propagation resistance. By traversing all adjacent cell pairs, including horizontal, vertical, and diagonal directions, a complete directed graph structure is established.

[0032] The identification of the force application initiation node is based on global temporal priority. The node with the earliest initial activation time and peak pressure exceeding 200 digital units among all sensing units is selected as the source node. Starting from the source node, an improved Dijkstra algorithm is used for path search. A priority queue is maintained to store nodes to be expanded. Each time, the node with the smallest current cumulative propagation cost is taken from the queue for expansion. For all outgoing neighbors of that node, if the cumulative cost to reach a neighbor through that node is less than the neighbor's recorded cost, the neighbor's cost is updated and added to the queue. This process is repeated until the force application termination node is reached. The termination node is defined as the last node activated in the edge region of the main path with a pressure decrease gradient greater than 50 units / second.

[0033] By tracing back the path with the minimum cumulative cost, the main path node sequence is obtained. The spatial coordinates of each node are calculated using the sensing unit number (i, j) and the grid spacing: x = i × 5 mm, y = j × 5 mm. The path nodes are arranged in ascending order of cumulative propagation cost, and a cubic spline curve is fitted using the least squares method. This curve is the axis of the main path, with the starting endpoint coordinates of the curve marked as (x0, y0). The pre-calibrated standard force application starting coordinates are (xs, ys), and the Euclidean distance between them is... Used as the starting offset.

[0034] The overall direction vector of the main path is determined by the endpoints of the curve. Let the endpoint of the curve be (x1, y1), then the direction vector is (x1-x0, y1-y0). The standard force application direction vector is preset according to the type of breech presentation. For complete breech presentation, the standard vector direction is from the lower right to the upper left of the abdomen, and after normalization, it is denoted as (vxs, vys). The actual direction vector is normalized to (vx, vy). The angle between the two vectors is calculated using the inverse cosine function. The included angle is the directional offset, measured in degrees. The initial offset and the directional offset are packaged into a two-dimensional feature vector [initial offset, directional offset] to form the path offset feature. This feature quantifies the degree of deviation between the actual applied force position and direction and the standard operating procedure, providing key input parameters for subsequent operational behavior evaluation.

[0035] Based on the pressure event stream, pressure peaks, pressure change slopes, and sensor unit activation sequences are extracted using an adaptive time window. These are then combined with the path offset features to generate operational behavior feature vectors. An operational behavior state transition model is constructed based on the operational behavior feature vector sequence within a continuous time window, including: A covariance matrix is ​​constructed for the pressure event stream and principal component analysis is performed. The time series of the first principal component is extracted and the envelope is calculated. The adaptive time window is defined by the time when the envelope enters the peak interval and the time when it leaves the peak interval. Within each adaptive time window, spatial interpolation is performed on the pressure peak of each sensing unit to generate a pressure distribution surface. The maximum modulus of the gradient field of the pressure distribution surface is calculated as the pressure change slope. Sensing units whose pressure peak exceeds the middle value of the window are identified and sorted by time to form a high-pressure activation sequence. The minimum convex hull is calculated based on the spatial coordinates of the sensing unit corresponding to the high-pressure activation sequence. The distance between the geometric center of the minimum convex hull and the predetermined standard center is used as the center offset, and the ratio of the area of ​​the minimum convex hull to the predetermined standard area is used as the coverage offset. This is then concatenated with the pressure change slope and the high-pressure activation sequence, and combined with the path offset features to generate an operation behavior feature vector. The operation behavior feature vectors of continuous time windows are used to construct a time series matrix. The Pearson correlation coefficient between feature vectors within the sliding window is calculated and the state boundaries are divided. The state representative vector of each state segment is calculated. The number of transitions and the transition time between the state representative vectors are statistically analyzed to construct the operation behavior state transition model.

[0036] A covariance matrix with dimensions n×n is established for the pressure values ​​of each sensing unit in the pressure event stream within a continuous sampling period. Principal component analysis is performed on this covariance matrix to extract the first principal component corresponding to the largest eigenvalue, resulting in a time series reflecting the overall pressure change trend. The envelope of this time series is calculated using Hilbert transform or moving average. The envelope value reflects the amplitude boundary of pressure changes. A peak threshold is set to 1.2 times the mean of the envelope. The moment when the envelope value first exceeds this threshold is recorded as the start point of the window, and the moment when the envelope value falls below the threshold is recorded as the end point of the window. This defines an adaptive time window, which automatically adapts to the differences in pressure change rates at different operational stages. The window is shorter during rapid force application and correspondingly longer during slow adjustment stages.

[0037] Within a defined adaptive time window, the pressure peak value of each sensing unit is extracted, and its spatial coordinates and pressure amplitude are recorded. Using Kriging interpolation or radial basis function interpolation, the pressure peak value data of discrete sensing units are extended into a continuous pressure distribution surface. The partial derivatives of this surface are calculated along the transverse and longitudinal directions of the abdominal zone, and the magnitude of the gradient field at each location is calculated. The maximum value of the gradient magnitude at all locations is taken as the pressure change slope, which reflects the degree of force concentration and the intensity of pressure change. The median of the pressure peak values ​​of all sensing units within the window is calculated, and sensing units whose pressure peak values ​​exceed the median are selected and arranged in ascending order according to the time when they reach the peak value to form a high-pressure activation sequence. This sequence records the spatiotemporal evolution order of the force application focus area.

[0038] Spatial coordinates (x, y) of each sensing unit in the high-voltage activation sequence i y i To perform convex hull calculation, the Graham scan algorithm is used to find the point with the smallest y-coordinate as the starting point. The polar angles of other points relative to the starting point are calculated and sorted by polar angle. Starting from the starting point, each point is examined sequentially, preserving the convexity of the convex hull. Finally, the smallest convex polygon containing all high-pressure points, i.e., the smallest convex hull H, is obtained. The geometric center coordinates (c...) of the smallest convex hull H are then calculated. x c y ), and the pre-calibrated standard operating center coordinates (s x , s y Calculate Euclidean distance The center offset D is obtained. The standard operating center coordinates are determined based on human anatomy and the external breech inversion procedure, and are usually located slightly above the center of the abdominal binder. The center offset reflects the degree of deviation of the applied force position from the ideal position during actual operation.

[0039] The area A_H of the minimum convex hull H is calculated, and its ratio R = A_H / A_S is taken with the standard operating coverage area A_S to obtain the coverage offset R. The standard operating coverage area is an ideal coverage range obtained based on statistics from a large number of successful cases, typically 30% to 40% of the total area of ​​the abdominal band. The coverage offset reflects the degree of diffusion or concentration of the applied force. An excessively large R value indicates that the applied force is too dispersed, while an excessively small R value indicates that the applied force is too concentrated, both of which will affect the operational effect. The center offset D, coverage offset R, pressure change slope G, high-pressure activation sequence encoding result A, and path offset feature P are concatenated in a fixed order to form the operational behavior feature vector V = [D, R, G, A, P]. Among them, the high-pressure activation sequence A is converted into a sequence code by arranging each sensing unit in a predetermined numbering order; the path offset feature P comes from the analysis results of the aforementioned pressure propagation path map, reflecting the difference between the pressure propagation trajectory and the standard path.

[0040] The operational behavior feature vectors generated by multiple consecutive time windows are arranged in chronological order to form a T×m dimensional temporal matrix, where T is the number of time windows. A sliding window of length k is set, and the Pearson correlation coefficient between adjacent feature vectors within the window is calculated. A state transition point is identified when the correlation coefficient is below 0.6, thus dividing the system into different state segments. The mean value of the feature vectors within each state segment is taken to obtain the state representative vector for that segment. The number of transitions between the state representative vectors is counted to establish a state transition frequency matrix. Simultaneously, the average duration of each state transition is recorded to form a state transition time matrix. After normalizing the state transition frequency matrix, a state transition probability matrix is ​​obtained. Combined with the state transition time matrix, this forms the operational behavior state transition model, which can characterize the evolutionary patterns and temporal dependencies of behavioral patterns during the operation process.

[0041] The operation behavior feature vectors within a continuous time window are used to construct a time series matrix. The Pearson correlation coefficient between feature vectors within the sliding window is calculated, and state boundaries are defined. The state representative vector for each state segment is calculated, and the number of transitions and transition times between state representative vectors are statistically analyzed to construct an operation behavior state transition model, including: The operation behavior feature vectors of continuous time windows are arranged in chronological order to form a time series matrix. A sliding window scan is performed on the time series matrix to calculate the Pearson correlation coefficient and construct a correlation coefficient matrix. The trace of the correlation coefficient matrix is ​​calculated to obtain the trace value time series. Change point detection is performed on the trace value time series, and the detected change point position is determined as the state boundary. The time series matrix is ​​segmented by the state boundary to obtain multiple state segments. The centroid of the operation behavior feature vector of each state segment is calculated as the state representative vector. The time span of each state segment is calculated as the state dwell time. Construct a state transition sequence, extract representative vectors of adjacent states in time sequence to form transition tuples, count the frequency of occurrence of the same transition tuple as the transition count, calculate the time span between the preceding state segment and the subsequent state segment corresponding to the same transition tuple and calculate the average value as the average transition time. State nodes are established using state representation vectors, with the state dwell time as a node attribute. Directed transition edges are established using transition tuples. The normalized transition rate is obtained by dividing the transition count by the state dwell time of the preceding state segment and used as the edge weight. The average transition time is used as the edge attribute. An operation behavior state transition model is constructed based on the state nodes and the directed transition edges.

[0042] After obtaining the operation behavior feature vectors for consecutive time windows, the operation behavior feature vectors corresponding to each time window are arranged in chronological order of acquisition time to form a time series matrix. Where T represents the total number of time windows and D represents the dimension of the operation behavior feature vector, a sliding window scan is performed on the time series matrix, the sliding window length is set to L and the step size is 1, and the submatrix is ​​extracted at the i-th position. For each eigenvector within a submatrix, calculate the Pearson correlation coefficient between any two eigenvectors to form an L×L correlation coefficient matrix. The Pearson correlation coefficient is calculated by the ratio of the covariance to the standard deviation of the eigenvectors. The diagonal elements of the correlation coefficient matrix are always 1, while the off-diagonal elements reflect the similarity between the eigenvectors.

[0043] The trace of the correlation coefficient matrix, which is the sum of its diagonal elements, is denoted as . As the sliding window moves across the time series matrix, the trace time series is obtained. When the operation behavior pattern changes, the similarity of feature vectors within the sliding window changes significantly, and the trace time series shows abrupt changes. Change point detection is performed on the trace time series. The cumulative sum control chart method is used to calculate the cumulative deviation and set the detection threshold. When the cumulative deviation exceeds the threshold, it is determined as a change point. The positions of each detected change point are mapped back to the time series matrix and determined as the state boundary. The time series matrix is ​​then divided into multiple state segments.

[0044] For each state segment, extract the feature vectors of all operational behaviors within that segment, calculate the arithmetic mean of each dimension, and obtain the centroid vector of that state segment, which serves as the state representative vector for that state. The state representative vector reflects the typical characteristics of operational behaviors within that state segment. At the same time, record the number of time windows contained in each state segment, multiply it by the duration of a single time window, and obtain the time span of that state segment, which serves as the state dwell time, reflecting the duration for which the operator remains in that state.

[0045] Based on the temporal order of state segments, extract the state representative vectors corresponding to adjacent state segments to form transition tuples. Each transition tuple represents the transition relationship from the preceding state to the subsequent state. Traverse all adjacent state segments and count the number of times the same transition tuple appears as a transition count, reflecting the frequency of the transition path. For each occurrence of the same transition tuple, calculate the time interval between the end time of the preceding state segment and the start time of the subsequent state segment. Calculate the arithmetic mean of all time intervals to obtain the average transition time, which characterizes the temporal characteristics of state transitions.

[0046] When establishing the operational behavior state transition model, each state representative vector is mapped to a state node in the graph. The state dwell time of that state is recorded as a node attribute, and transition tuples are mapped to directed transition edges, with the starting point of the edge being the preceding state node and the ending point being the subsequent state node. The normalized transition rate is calculated as the edge weight, specifically by dividing the transition count by the state dwell time of the preceding state segment to eliminate the influence of differences in dwell times between different states, making the transition rates comparable. The average transition time is recorded as an edge attribute, reflecting the average delay of the transition. Based on the state nodes, directed transition edges, and their attributes, a directed weighted graph is constructed as the operational behavior state transition model. This model can characterize the evolution of operational behavior between different states, providing a basis for state transitions in subsequent predictions and risk assessments.

[0047] Based on the state transition model, subsequent feature vectors are extrapolated and predicted. A risk assessment result is generated by combining the prediction confidence level and the stress safety upper limit. This risk assessment result, along with the stress event stream, is then synchronously uploaded to the remote monitoring platform, including: Extract the current state node from the state transition model, trace back the historical transition path to identify the cyclic transition sub-path, and determine whether the current state node is located within the cyclic transition sub-path. If it is located within the cyclic, extract the state representative vectors of each state node in the cyclic to form a cyclic feature sequence, and periodically extrapolate the cyclic feature sequence to obtain the predicted feature vector. If it is not located within the cyclic, extract the state representative vector of the successor state node with the largest normalized transition rate as the predicted feature vector. For cyclic transfer prediction, the ratio of the number of repeated cycles to the total number of historical transfers is used as the prediction confidence level; for non-cyclic transfer prediction, the difference between the maximum and the second largest normalized transfer rate is used as the prediction confidence level. The pressure component is extracted from the predicted feature vector, the difference between the pressure component and the pressure safety upper limit is calculated to obtain the pressure over-limit value, the pressure over-limit value is multiplied by the prediction confidence level to obtain the confidence weighted over-limit value, and the confidence weighted over-limit value is mapped to a preset risk partition to obtain the risk assessment result. The upload time interval is determined based on the risk assessment results. The stress event stream is sampled according to the upload time interval. The risk assessment results are appended to the sampled stress event stream and uploaded to the remote monitoring platform.

[0048] After the pressure sensing layer embedded in the abdominal band collects real-time pressure signals and timestamps from each sensing unit, the operational behavior state transition model has been constructed. The current state node is extracted from this model, and the historical state transition path before this node is traced back. By analyzing the recurrence of state nodes in the historical transition path, the existence of cyclic transition sub-paths is identified. Specifically, each state node in the historical path is hashed, and the position index of the same code value is counted. If at least two state node sequences with the same code value reappear at intervals, the sequence is determined to be a cyclic transition sub-path. Whether the current state node is located within an identified cyclic transition sub-path is determined by comparing the code value of the current state node with the code value of any node in the cycle.

[0049] If the current state node is located within a cyclic transition sub-path, extract the state representative vectors corresponding to all state nodes in the cycle. Each state representative vector contains operational behavior feature components such as the pressure peak, pressure change slope, and sensor unit activation sequence corresponding to that state node. Arrange the extracted state representative vectors according to their temporal order in the cycle to form a cyclic feature sequence. Based on the periodicity of the cycle, extrapolate the cyclic feature sequence chronologically to obtain the prediction feature vector corresponding to the prediction time. For example, if the cycle contains 5 state nodes, and the current state node is the 3rd node in the cycle, then the feature vector for predicting the next time step is the state representative vector of the 4th node in the cycle.

[0050] If the current state node is not within the cyclic transition sub-path, search for all successor state nodes pointed to by the current state node in the state transition model, calculate the transition rate for each successor state node. The transition rate is defined as the ratio of the number of transitions from the current state node to the successor state node to the total number of transitions originating from the current state node. Normalize all transition rates to ensure that the sum of the transition rates is 1. Select the successor state node with the largest normalized transition rate and extract the state representative vector of the successor state node as the prediction feature vector.

[0051] For the predicted feature vector obtained based on cyclic transition prediction, the number of times the cycle repeats in the historical state transition path is counted, and this number is divided by the total number of historical state transitions to obtain the frequency ratio of the cycle recurrence. This ratio is used as the prediction confidence. For the predicted feature vector obtained based on non-cyclic transition prediction, the maximum and second largest normalized transition rates are extracted, and the difference between the two is calculated as the prediction confidence. The larger the difference, the higher the certainty of the selection of the optimal successor state node and the more reliable the prediction result.

[0052] Pressure-related components, including the predicted pressure peak, are extracted from the predicted feature vector. This pressure component is compared to a pre-defined safe pressure upper limit, and the difference between the two is calculated to obtain the pressure exceedance value. If the pressure component does not exceed the safe upper limit, the pressure exceedance value is 0; otherwise, it is a positive value. The pressure exceedance value is multiplied by the corresponding prediction confidence level to obtain the confidence-weighted exceedance value. This confidence-weighted exceedance value comprehensively considers both the degree of pressure exceedance and the reliability of the prediction result.

[0053] Pre-defined boundary thresholds for risk zones are used to map confidence-weighted out-of-limit values ​​to the corresponding risk zones. Risk zones can be divided into three levels: low risk, medium risk, and high risk. If the confidence-weighted out-of-limit value is below the first threshold, it is mapped to low risk; if it is between the first and second thresholds, it is mapped to medium risk; and if it is above the second threshold, it is mapped to high risk. The risk level obtained from the mapping is the risk assessment result.

[0054] The upload interval is determined based on the risk assessment results. The higher the risk level, the shorter the upload interval. For high-risk situations, the upload interval is set to 0.5 seconds; for medium-risk situations, it is set to 2 seconds; and for low-risk situations, it is set to 5 seconds. The pressure event stream is sampled according to the determined upload interval, and the pressure signal, timestamp, and location identifier corresponding to the sampling time are extracted. The risk assessment results are encapsulated as additional information along with the sampled pressure event stream and uploaded to the remote monitoring platform via a wireless communication module, achieving synchronous transmission of pressure data and risk assessment results. In the clinical application of external breech presentation, to ensure real-time monitoring, the end-to-end latency of the communication link is controlled within 200 milliseconds.

[0055] After receiving data packets, the remote monitoring platform verifies data integrity and parses the risk assessment results. Combining historical data with the current operational stage, it performs risk trend analysis using a sliding window method to calculate the rate of change and volatility of recent risk scores, identifying patterns such as rapid risk escalation or sustained high risk levels. Based on the risk assessment results and trend analysis, the remote monitoring platform generates corresponding operational intervention instructions. These instructions are categorized into four types: information prompts, operational suggestions, warning adjustments, and emergency interventions, each corresponding to different risk levels and trend characteristics.

[0056] Information-based instructions are used when the risk is in the safe zone or has just entered the warning zone. They mainly provide real-time feedback on operating parameters and status, such as "Current pressure is within the ideal range" and "Please maintain the stability of the current technique." Operational suggestion instructions are applicable when the risk is in the upper part of the warning zone or is trending upwards. They provide specific suggestions for optimizing the operation, such as "Slightly reduce the pressure on the right side" and "You can moderately increase the operating speed." Warning and adjustment instructions are for situations where the risk has reached the warning zone. They clearly indicate the direction and magnitude of the adjustment needed, such as "Please move the application point two centimeters to the left" and "Immediately reduce the pressure intensity by 20%." Emergency intervention instructions are used when the risk enters the danger zone or rapidly rises to the upper limit of the warning zone. They provide clear instructions for interruption and reset, such as "Immediately stop the operation and release all pressure" and "Reposition the abdominal binder according to standard procedures."

[0057] After the intervention command is generated, the remote monitoring platform encodes it into a standard command format, which includes fields such as command type identifier, specific operation content, priority marker and timestamp. High-priority commands (such as emergency intervention commands) are given a special identifier to ensure that they are displayed and processed first on the operation terminal. The command is sent to the operation terminal through a dedicated command channel. This channel adopts a redundant transmission mechanism to ensure that the command is reliably delivered under various network conditions. To cope with extreme situations, a minimum necessary command set is designed so that critical intervention information can be transmitted even under extremely low bandwidth.

[0058] After receiving an intervention command, the operating terminal presents it to the operator through appropriate interface elements according to the command type and priority. Information prompts are displayed as status bar messages; operation suggestion commands are displayed through a semi-transparent floating window; warning and adjustment commands use eye-catching pop-ups and sound prompts; and emergency intervention commands draw immediate attention through full-screen warnings and vibration feedback. The command display adopts an intuitive visualization method, such as using arrows to indicate the direction of movement and using color gradients to indicate the pressure adjustment range, to help the operator quickly understand and execute the command.

[0059] This invention also provides a system for acquiring and remotely monitoring abdominal binder pressure data in breech presentation, comprising: The pressure acquisition unit is used to acquire real-time pressure signals and timestamps from each sensing unit during the external breech reversal operation by means of a pressure sensing layer embedded in the abdominal binder, forming a pressure event stream with time sequence annotations and location identifiers; The path construction unit is used to construct a pressure propagation path map representing the direction of force application based on the pressure transmission timing and amplitude differences between adjacent sensing units, and to calculate the path offset characteristics. The feature extraction unit is used to extract pressure peaks, pressure change slopes and sensing unit activation sequences based on the pressure event stream using an adaptive time window, generate operation behavior feature vectors by combining the path offset features, and construct an operation behavior state transition model based on the operation behavior feature vector sequence of continuous time windows. The risk assessment unit is used to extrapolate and predict subsequent feature vectors based on the state transition model, generate risk assessment results by combining the prediction confidence and the pressure safety upper limit, and synchronously upload the risk assessment results and the pressure event stream to the remote monitoring platform. The instruction generation unit is used by the remote monitoring platform to generate operation intervention instructions based on the risk assessment results and send them to the operation terminal.

[0060] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0061] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0062] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for collecting and remotely monitoring abdominal binder pressure data in breech position with external rotation, characterized in that, include: By embedding a pressure sensing layer in the abdominal binder, real-time pressure signals and timestamps of each sensing unit during the external breech inversion operation are collected, forming a pressure event stream with time sequence annotation and location identifier; Based on the pressure transmission timing and amplitude differences between adjacent sensing units, a pressure propagation path map representing the direction of force application is constructed, and the path offset characteristics are calculated. Based on the pressure event stream, pressure peak, pressure change slope and sensor unit activation sequence are extracted using an adaptive time window. Combined with the path offset features, an operation behavior feature vector is generated. An operation behavior state transition model is constructed based on the operation behavior feature vector sequence of continuous time windows. The subsequent feature vectors are extrapolated and predicted based on the state transition model. A risk assessment result is generated by combining the prediction confidence and the pressure safety limit. The risk assessment result and the pressure event stream are synchronously uploaded to the remote monitoring platform. The remote monitoring platform generates an operation intervention command based on the risk assessment result and sends it to the operation terminal.

2. The method according to claim 1, characterized in that, By embedding a pressure-sensing layer within the abdominal binder, real-time pressure signals and timestamps from each sensing unit during external breech inversion are collected, forming a pressure event stream with time-series and location markers, including: Based on the abdominal anatomical structure, the sensing units in the pressure sensing layer are divided into zones and classified, and each sensing unit is assigned a unique location identifier corresponding to its anatomical location. The average pressure of each sensing unit at rest before force application is used as the individualized baseline. During the acquisition process, the baseline difference operation is continuously performed on the real-time pressure signal to extract the effective force application increment signal after removing the static pre-pressure. The event trigger boundary is determined based on the amplitude change characteristics of the effective force increment signal. The effective acquisition window is dynamically captured with the event trigger boundary as the start and end time. Within the effective acquisition window, the timestamps of all sensing units are aligned with a unified time base to eliminate multi-channel acquisition timing deviation. The effective force increment signal amplitude of each sensing unit within the effective acquisition window is combined with the alignment timestamp, so that each event entry carries a location identifier, effective force increment amplitude and alignment timestamp simultaneously, forming a pressure event stream with time sequence labeling and location labeling.

3. The method according to claim 1, characterized in that, Based on the differences in pressure transmission timing and amplitude between adjacent sensing units, a pressure propagation path map representing the direction of force application is constructed, and path offset characteristics are calculated, including: Based on the pressure event flow, the first activation time of the pressure signal of each sensing unit is extracted. The direction of the directed edge propagation is determined by the positive and negative directions of the difference between the first activation times of adjacent sensing units. The sum of the absolute values ​​of the difference between the first activation times and the absolute values ​​of the peak amplitude is used as the propagation cost of the directed edge. The directed edges with each sensing unit as a node and the propagation cost and propagation direction as attributes are connected to adjacent nodes to form a pressure propagation path graph. Based on the pressure propagation path diagram, starting from the force application node, nodes are gradually expanded from small to large while maintaining path connectivity. The path with the minimum cumulative propagation cost to the force termination node is extracted as the main pressure propagation path. The corresponding spatial coordinates of each node on the main path are retrieved by the position identifiers of each node, arranged in the order of cumulative propagation cost, and then fitted to obtain the main path axis. The distance between the starting endpoint coordinates of the main path axis and the pre-calibrated standard force application starting coordinates is used as the starting offset of the path offset feature. The angle between the overall direction vector of the main path axis and the pre-calibrated standard force application direction vector is used as the directional offset of the path offset feature. The starting offset and the directional offset are combined to obtain the path offset feature.

4. The method according to claim 1, characterized in that, Based on the pressure event stream, pressure peaks, pressure change slopes, and sensor unit activation sequences are extracted using an adaptive time window. These are then combined with the path offset features to generate operational behavior feature vectors. An operational behavior state transition model is constructed based on the operational behavior feature vector sequence within a continuous time window, including: A covariance matrix is ​​constructed for the pressure event stream and principal component analysis is performed. The time series of the first principal component is extracted and the envelope is calculated. The adaptive time window is defined by the time when the envelope enters the peak interval and the time when it leaves the peak interval. Within each adaptive time window, spatial interpolation is performed on the pressure peak of each sensing unit to generate a pressure distribution surface. The maximum modulus of the gradient field of the pressure distribution surface is calculated as the pressure change slope. Sensing units whose pressure peak exceeds the middle value of the window are identified and sorted by time to form a high-pressure activation sequence. The minimum convex hull is calculated based on the spatial coordinates of the sensing unit corresponding to the high-pressure activation sequence. The distance between the geometric center of the minimum convex hull and the predetermined standard center is used as the center offset, and the ratio of the area of ​​the minimum convex hull to the predetermined standard area is used as the coverage offset. This is then concatenated with the pressure change slope and the high-pressure activation sequence, and combined with the path offset features to generate an operation behavior feature vector. The operation behavior feature vectors of continuous time windows are used to construct a time series matrix. The Pearson correlation coefficient between feature vectors within the sliding window is calculated and the state boundaries are divided. The state representative vector of each state segment is calculated. The number of transitions and the transition time between the state representative vectors are statistically analyzed to construct the operation behavior state transition model.

5. The method according to claim 4, characterized in that, The operation behavior feature vectors within a continuous time window are used to construct a time series matrix. The Pearson correlation coefficient between feature vectors within the sliding window is calculated, and state boundaries are defined. The state representative vector for each state segment is calculated, and the number of transitions and transition times between state representative vectors are statistically analyzed to construct an operation behavior state transition model, including: The operation behavior feature vectors of continuous time windows are arranged in chronological order to form a time series matrix. A sliding window scan is performed on the time series matrix to calculate the Pearson correlation coefficient and construct a correlation coefficient matrix. The trace of the correlation coefficient matrix is ​​calculated to obtain the trace value time series. Change point detection is performed on the trace value time series, and the detected change point position is determined as the state boundary. The time series matrix is ​​segmented by the state boundary to obtain multiple state segments. The centroid of the operation behavior feature vector of each state segment is calculated as the state representative vector. The time span of each state segment is calculated as the state dwell time. Construct a state transition sequence, extract representative vectors of adjacent states in time sequence to form transition tuples, count the frequency of occurrence of the same transition tuple as the transition count, calculate the time span between the preceding state segment and the subsequent state segment corresponding to the same transition tuple and calculate the average value as the average transition time. State nodes are established using state representation vectors, with the state dwell time as a node attribute. Directed transition edges are established using transition tuples. The normalized transition rate is obtained by dividing the transition count by the state dwell time of the preceding state segment and used as the edge weight. The average transition time is used as the edge attribute. An operation behavior state transition model is constructed based on the state nodes and the directed transition edges.

6. The method according to claim 1, characterized in that, Based on the state transition model, subsequent feature vectors are extrapolated and predicted. A risk assessment result is generated by combining the prediction confidence level and the stress safety upper limit. This risk assessment result, along with the stress event stream, is then synchronously uploaded to the remote monitoring platform, including: Extract the current state node from the state transition model, trace back the historical transition path to identify the cyclic transition sub-path, and determine whether the current state node is located within the cyclic transition sub-path. If it is located within the cyclic, extract the state representative vectors of each state node in the cyclic to form a cyclic feature sequence, and periodically extrapolate the cyclic feature sequence to obtain the predicted feature vector. If it is not located within the cyclic, extract the state representative vector of the successor state node with the largest normalized transition rate as the predicted feature vector. For cyclic transfer prediction, the ratio of the number of repeated cycles to the total number of historical transfers is used as the prediction confidence level; for non-cyclic transfer prediction, the difference between the maximum and the second largest normalized transfer rate is used as the prediction confidence level. The pressure component is extracted from the predicted feature vector, the difference between the pressure component and the pressure safety upper limit is calculated to obtain the pressure over-limit value, the pressure over-limit value is multiplied by the prediction confidence level to obtain the confidence weighted over-limit value, and the confidence weighted over-limit value is mapped to a preset risk partition to obtain the risk assessment result. The upload time interval is determined based on the risk assessment results. The stress event stream is sampled according to the upload time interval. The risk assessment results are appended to the sampled stress event stream and uploaded to the remote monitoring platform.

7. A system for acquiring and remotely monitoring abdominal binder pressure data in breech position, used to implement the method as described in any one of claims 1-6, characterized in that, include: The pressure acquisition unit is used to acquire real-time pressure signals and timestamps from each sensing unit during the external breech reversal operation by means of a pressure sensing layer embedded in the abdominal binder, forming a pressure event stream with time sequence annotations and location identifiers; The path construction unit is used to construct a pressure propagation path map representing the direction of force application based on the pressure transmission timing and amplitude differences between adjacent sensing units, and to calculate the path offset characteristics. The feature extraction unit is used to extract pressure peaks, pressure change slopes and sensing unit activation sequences based on the pressure event stream using an adaptive time window, generate operation behavior feature vectors by combining the path offset features, and construct an operation behavior state transition model based on the operation behavior feature vector sequence of continuous time windows. The risk assessment unit is used to extrapolate and predict subsequent feature vectors based on the state transition model, generate risk assessment results by combining the prediction confidence and the pressure safety upper limit, and synchronously upload the risk assessment results and the pressure event stream to the remote monitoring platform. The instruction generation unit is used by the remote monitoring platform to generate operation intervention instructions based on the risk assessment results and send them to the operation terminal.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images