A blood pressure dynamic evaluation method and device based on a body position conversion stage

Abstract

This invention discloses a method and device for dynamic blood pressure assessment based on postural transition phases. It identifies postural transitions from postural sensor data and delineates the boundaries of the preparation, transition, and stabilization phases based on individualized reflex delay compensation. It constructs an individualized blood pressure baseline curve based on phased sampling and establishes a phased reference benchmark. It detects a continuous downward trend in the blood pressure deviation sequence and applies individualized rate constraints to form a dynamic trajectory. It removes physiological periodic components using an adaptive rhythm template and determines abnormal fluctuation ranges and completes phased risk assessment based on a joint score of non-rhythmic fluctuation amplitude and duration. It integrates the risk assessment results with the individual baseline level for adaptive weighting based on postural orientation, determines differentiated warning thresholds, and outputs blood pressure change assessment results. This achieves individualized and phased-differentiated accurate assessment of blood pressure abnormalities induced by postural transitions, improving the accuracy and adaptability of orthostatic hypotension risk assessment.

Description

Technical Field

[0001] This invention relates to the field of medical and health monitoring technology, and in particular to a method and device for dynamic blood pressure assessment based on body position transitions. Background Technology

[0002] During the process of changing from a lying to a standing or vice versa, the change in the direction of gravity causes a redistribution of blood in the body. The baroreceptor regulatory system needs to complete the dynamic readjustment of vascular tone within a short period of time. For the elderly, patients with hypertension, and those with impaired autonomic nervous system function, this regulatory process is characterized by slow response and insufficient regulatory capacity. Blood pressure can fluctuate significantly within seconds to tens of seconds after the change of position, which can induce adverse events such as dizziness, fainting, or even falls in severe cases. Orthostatic hypotension has been recognized as one of the independent risk factors for cardiovascular and cerebrovascular accidents in the above-mentioned groups.

[0003] Existing dynamic blood pressure monitoring methods mostly collect blood pressure values ​​at fixed time intervals and simply compare them with static reference values, failing to correlate blood pressure changes with specific postural transitions, making it difficult to accurately define the start and end times of the regulatory response window. Furthermore, significant differences exist in the delay of autonomic reflexes among individuals, and using uniform time boundaries to divide the regulatory phase often leads to missed or misjudged critical abnormal periods. In addition, the periodic blood pressure fluctuations caused by respiratory movements and slow-rhythmic vasomotor activity are highly overlapping with the abnormal shifts induced by postural transitions in the time domain. Existing methods lack effective separation mechanisms for these two components, making risk assessment results highly susceptible to interference from physiological background fluctuations, making it difficult to guarantee the individual adaptability of warning thresholds, and failing to distinguish the asymmetric differences in blood pressure regulation sensitivity between supine to standing and standing to supine positions. Summary of the Invention

[0004] This invention discloses a method and device for dynamic blood pressure assessment based on the body position transition phase. It aims to accurately delineate the boundary of the regulation phase through individualized reflex delay compensation, and integrates phased baseline construction, rate constraint deviation analysis and adaptive rhythm separation to conduct phase-differentiated risk assessment of blood pressure abnormalities induced by body position transition. Furthermore, it adaptively determines differentiated warning thresholds by combining the characteristics of the body position transition direction and the individual baseline level, and finally achieves individualized and accurate judgment and assessment output of the risk of orthostatic hypotension, providing quantitative support for early identification and individualized intervention decisions for orthostatic hypotension.

[0005] The first aspect of this invention proposes a method for dynamic blood pressure assessment based on body position transition phases, comprising the following steps: Collect patient posture sensor data and blood pressure sensor data, identify posture transition events by performing posture transition recognition on the posture sensor data, and divide and identify the response lag compensation generation stage based on the posture transition events. The blood pressure sensor data is sampled in stages using the stage division identifier to generate a multi-stage blood pressure sampling sequence. An individualized baseline is constructed from the multi-stage blood pressure sampling sequence to generate a blood pressure reference curve. The blood pressure reference curve is compared and mapped with the real-time blood pressure sampling value to form a stage reference benchmark. Based on the stage reference benchmark, the deviation of the real-time blood pressure sampling value is calculated point by point to generate a blood pressure deviation sequence. The continuous downward trend and recovery characteristics after the decline are detected in the blood pressure deviation sequence to generate rate constraint parameters. Based on the rate constraint parameters, the blood pressure deviation sequence is rate-constrained to form a dynamic blood pressure change trajectory. The abnormal fluctuation range is determined by separating the physiological rhythm fluctuations of the blood pressure dynamic change trajectory, and a staged risk assessment map is formed based on the abnormal fluctuation range of the blood pressure dynamic change trajectory. Based on the stage risk assessment map and the blood pressure baseline curve, a differentiated early warning threshold is determined by adaptive fusion of body position. Based on the differentiated early warning threshold, the blood pressure change assessment result is output when the threshold is exceeded.

[0006] A second aspect of the present invention provides a dynamic blood pressure assessment device based on the body position transition phase, comprising: The data acquisition module is used to collect patient posture sensor data and blood pressure sensor data, identify posture transition events by performing posture transition identification on the posture sensor data, and perform response lag compensation generation stage segmentation based on the posture transition events. The baseline establishment module is used to sample the blood pressure sensing data in stages according to the stage division identifier to generate a multi-stage blood pressure sampling sequence, perform individualized baseline construction on the multi-stage blood pressure sampling sequence to generate a blood pressure baseline curve, and establish a stage comparison mapping between the blood pressure baseline curve and the real-time blood pressure sampling value to form a stage reference baseline. The deviation tracking module is used to calculate the deviation of the real-time blood pressure sampling value point by point according to the stage reference benchmark to generate a blood pressure deviation sequence, detect the continuous downward trend and recovery characteristics after the decline of the blood pressure deviation sequence to generate rate constraint parameters, and perform rate constraint on the blood pressure deviation sequence according to the rate constraint parameters to form a dynamic blood pressure change trajectory. The risk assessment module is used to separate the physiological rhythm fluctuations of the blood pressure dynamic change trajectory to determine the abnormal fluctuation range, and to perform risk assessment on the blood pressure dynamic change trajectory based on the abnormal fluctuation range to form a stage risk assessment map. The early warning output module is used to determine a differentiated early warning threshold by adaptively fusing the stage risk assessment map and the blood pressure baseline curve according to body position, and to output the blood pressure change assessment result based on the differentiated early warning threshold to determine if the threshold is exceeded.

[0007] The beneficial effects of this invention are reflected in the following aspects: First, by performing individualized delay compensation based on historical response data at the time of body position transition, the problem of mislabeling of the regulation phase caused by uniform time boundaries is solved, making the division of the preparation phase, transition phase, and stabilization phase consistent with the actual temporal sequence of the patient's autonomic nerve reflexes. On this basis, blood pressure sampling data is subjected to intra-phase stability screening, outlier exclusion, and segmented baseline fitting to construct a blood pressure reference baseline that closely reflects the individual's physiological state. Weighted comparison mapping is generated through segmented variability assessment, providing an individualized stage reference basis for subsequent deviation analysis. Second, by performing continuous downward trend detection and trough-recovery inflection point localization on the blood pressure deviation sequence, individualized rate characteristics of the blood pressure drop-recovery process induced by body position transition are extracted. Based on this, segmented rate constraint parameters with differentiated drop and recovery directions are set, and point-by-point rate constraints are applied to the deviation sequence. While preserving the clinically significant rapid downward trend, abnormal jumps caused by sampling noise are eliminated, forming a trajectory that can truly reflect the dynamic regulation process of blood pressure. Finally, after stripping away the gradual trend by using the moving average, individualized dual-band rhythmic period estimation was performed on the residual sequence, and an adaptive rhythmic template was constructed to separate physiological periodic components such as respiration and Mayer waves from blood pressure fluctuations. Abnormal fluctuation intervals were determined based on the joint scoring of the amplitude and duration of non-rhythmic components, and a phased risk assessment was completed. Furthermore, by analyzing the risk gradient between stages, the characteristics of the postural transition direction were extracted, and adaptive fusion weights were generated by combining the directional sensitivity of the individual baseline level. Finally, differentiated warning thresholds for each stage and direction were determined, improving the adaptability of postural hypotension risk assessment to individual differences and asymmetry in postural transition direction. Attached Figure Description

[0008] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0009] Figure 1 This is a flowchart illustrating a method for dynamic blood pressure assessment based on body position transitions according to the present invention.

[0010] Figure 2 This is a structural block diagram of a blood pressure dynamic assessment device based on the body position transition stage according to the present invention. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0013] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0014] The technical solutions of the embodiments of this application will be described below.

[0015] like Figure 1 As shown, this embodiment of the invention provides a method for dynamic blood pressure assessment based on the body position transition phase, including the following steps S110-S150: Step S110: Collect patient position sensor data and blood pressure sensor data, identify position transition events by performing position transition identification on the position sensor data, and divide and mark the response lag compensation generation stage based on the position transition events.

[0016] Specifically, patient posture and blood pressure data are collected. Posture data is acquired via a triaxial accelerometer integrated into the patient's abdominal binder. The binder is wrapped around and fixed to the patient's abdomen, ensuring a stable fit between the sensor and the torso. The sensor sampling frequency is set to 50Hz, outputting X, Y, and Z-axis acceleration components. The resultant force vector of the three axes is used to continuously calculate the torso elevation angle θ. The formula for calculating the torso elevation angle θ is... Where a_x, a_y, and a_z are the acceleration components along the X, Y, and Z axes, respectively, and θ is in degrees, ranging from −90 degrees to 90 degrees. Blood pressure data is acquired by a continuous non-invasive blood pressure sensor in the radial artery at a sampling frequency of 100 Hz, outputting stroke-by-stroke systolic and diastolic blood pressure values. The timestamps of the blood pressure data and the body position data are aligned via a hardware synchronization module. During acquisition, the body position data undergoes low-pass filtering at a cutoff frequency of 2 Hz to eliminate motion artifacts. The filtered triaxial acceleration components are used for the stable calculation of the trunk elevation angle, with an accuracy better than 1 degree. The stable fixation structure of the abdominal binder effectively suppresses the relative displacement of the sensor during the patient's daily movements, reducing interference from non-positional transitions on the elevation angle calculation. When the sampled value of the blood pressure data exceeds the range of 60 to 180 mmHg for systolic blood pressure or 40 to 110 mmHg for diastolic blood pressure, the sensor hardware adds an invalid marker to the corresponding sampling point. The invalid marker distinguishes between valid sampling points and hardware-abnormal sampling points.

[0017] Postural transition events are identified by analyzing postural sensor data. A trigger condition is met when the trunk elevation angle changes by more than 30 degrees and the new position is maintained for more than 3 seconds after the change. The algorithm uses a sliding window with a window length of 5 seconds and a step size of 0.1 seconds to check frame-by-frame whether the elevation angle time series of the postural sensor data meets this condition. The trigger time for a postural transition event is when the trunk elevation angle in the postural sensor data crosses the 45-degree dividing point. 45 degrees, as the geometric midpoint threshold between supine and upright positions, has good temporal representativeness in clinical practice. The polarity of the elevation angle change in the postural sensor data determines the direction of the postural transition event: a monotonically increasing elevation angle exceeding 30 degrees corresponds to a transition from supine to standing, and a monotonically decreasing elevation angle exceeding 30 degrees corresponds to a transition from standing to supine. When a patient quickly rises from a supine position, the elevation angle of the position sensor data increases from 8 degrees to 82 degrees within approximately 2.5 seconds. The trigger time for the position transition event is recorded as the moment the elevation angle crosses the 45-degree boundary, and the transition direction is marked as from supine to standing. The sliding window detection of the position sensor data operates in real-time during acquisition, synchronously outputting the current trigger status within each window step. Once the trigger condition is met for the first time, the trigger time of the position transition event is locked to prevent the same position transition action from being recorded repeatedly. The position transition event includes the trigger time and the transition direction. The trigger time is recorded with millisecond precision in the timestamp coordinate system of the position sensor data, and the transition direction is either from supine to standing or from standing to supine.

[0018] In some embodiments, the step of generating stage division identifiers based on response hysteresis compensation for the body position transition event includes: generating a baroreceptor reflection delay window by estimating the response delay based on the body position transition event; generating a calibration delay parameter by performing individualized calibration of the baroreceptor reflection delay window; generating a hysteresis compensation stage boundary by using the calibration delay parameter to perform time compensation on the body position transition event; and determining the stage division identifier by dividing the preparation period, transition period, and stabilization period according to the hysteresis compensation stage boundary.

[0019] A baroreceptor reflex delay window is generated based on response delay estimation using a postural transition event. The baroreceptor reflex is a core physiological mechanism for automatic blood pressure regulation after a postural change. The carotid sinus and aortic arch receptors detect blood pressure changes within hundreds of milliseconds after a positional change, and the reflex arc is transmitted via the vagus and sympathetic nerve pathways. The overall response exhibits an inherent neural conduction delay. The trigger moment of the postural transition event is used as the time reference for response delay estimation. In the initial stage starting from this moment, blood pressure is still in a state of inertial fluctuation, and active baroreceptor regulation has not yet intervened. Directly using the trigger moment of the postural transition event as the starting point of this stage would lead to mislabeling of the regulatory transition period. The initial offset of the baroreceptor reflex delay window is set to 1.5 seconds after the trigger moment of the postural transition event; this value is determined based on the average delay time in a healthy adult population. The direction of the postural transition event contributes to the construction of the baroreceptor reflex delay window. The baroreceptor reflex delay window for transitioning from a supine to an upright position is set to 8 seconds, while the delay window for transitioning from an upright to a supine position is set to 6 seconds. The difference in duration stems from the asymmetry of the gravity-assisted effect in the two directions. When a patient lies down from a standing position, venous return increases, and blood pressure gradually stabilizes within 6 seconds. However, when rising from a supine position, gravity causes blood pooling in the lower limbs, requiring the baroreceptor to complete vascular tone readjustment for a longer period. The 8-second window duration matches the adjustment time in this direction. The baroreceptor reflex delay window consists of two parameters: window initial offset and window duration. Both parameters are measured based on the trigger time of the postural transition event.

[0020] Individualized calibration of the baroreceptor reflex delay window generates calibration delay parameters. The population default value for the baroreceptor reflex delay window is determined based on statistics from healthy adults, without considering individual delay differences in elderly patients and patients with autonomic dysfunction. The calibration process extracts the measured delay between the blood pressure response initiation time and the postural change event trigger time from the patient's historical postural change records. The median of the measured delays from at least three effective postural changes is used as the individualized starting offset. The individualized correction of the baroreceptor reflex delay window duration is determined based on the 75th percentile of the time required for blood pressure to recover to steady state in historical records. When the difference between the default starting offset of the baroreceptor reflex delay window and the patient's measured median delay exceeds 0.5 seconds, the calibrated starting offset of the delay parameter is based on the individualized measured value; if the difference is less than 0.5 seconds, the default starting offset is retained. When the difference between the default window duration of the baroreceptor reflex delay window and the 75th percentile of the time required for blood pressure to return to steady state in the patient's historical records exceeds 1.5 seconds, the calibrated window duration of the delay parameter is based on the individualized 75th percentile value; if the difference is less than 1.5 seconds, the default window duration is retained. The calibration delay parameter includes the calibrated starting offset and the calibrated window duration, both of which are determined by correcting the corresponding parameters of the baroreceptor reflex delay window using historical data. Historically measured delays in elderly patients with impaired autonomic function often reach 2.8 seconds; therefore, the calibration delay parameter corrects the starting offset from the default 1.5 seconds to 2.8 seconds.

[0021] The calibration delay parameter is used to compensate for the timing of postural transition events, generating the lag compensation phase boundary. If the original trigger time T0 of the postural transition event is directly used as the start time of the transition period, the inertial fluctuation period between T0 and the actual initiation time of blood pressure regulation will be mistakenly included in the transition period. The calibrated start offset of the calibration delay parameter provides the time correction required to eliminate this bias. In elderly patients with an autonomic reflex delay of up to 2.8 seconds, the calibration delay parameter shifts the start time of the transition period from T0 to T0 plus 2.8 seconds, aligning the phase boundary with the actual initiation time of blood pressure regulation. The preparation period termination time of the lag compensation phase boundary is determined by adding the calibrated start offset of the calibration delay parameter to the trigger time of the postural transition event; this time corresponds to the physiological starting point where the baroreceptor regulation mechanism begins to intervene. The calibrated window duration of the calibration delay parameter determines the duration of the transition period. The start time of the transition period is directly connected to the preparation period termination time, and the termination time of the transition period is obtained by superimposing the preparation period termination time with the calibrated window duration. The direction of the positional transition event and the calibration delay parameter together determine the final value of the hysteresis compensation phase boundary. The post-calibration window duration of the calibration delay parameter during the transition from supine to upright position is usually longer than the corresponding value during the transition from upright to supine position, resulting in a systematic difference in the boundary position between the two transition directions. The hysteresis compensation phase boundary includes the end time of the preparation period and the end time of the transition period. The end time of the transition period is the beginning time of the stabilization period. These two times, arranged sequentially, form a complete phase boundary sequence.

[0022] The preparation, transition, and stabilization phases are defined based on the lag compensation phase boundaries, with stage identifiers used to determine these phases. The preparation phase, defined by the termination time of the preparation phase at the lag compensation phase boundary, covers the period before the patient completes a positional change, during which blood pressure fluctuates from baseline. The transition phase, defined by both the termination times of the preparation and transition phases at the lag compensation phase boundary, is characterized by dynamic changes in blood pressure due to both gravity redistribution and baroreceptor reflex activation, representing a high-risk period for orthostatic hypotension. The stabilization phase extends from the termination time of the transition phase at the lag compensation phase boundary to the end of the data retention window, corresponding to the period when blood pressure re-establishes a steady state under the new position. Stage identifiers assign a stage type label to each sampling point of the blood pressure sensor data: a preparation phase label is used when the sampling point's timestamp falls within the preparation phase, a transition phase label when it falls within the transition phase, and a stabilization phase label when it falls within the stabilization phase. The phase division identifier includes a phase type label sequence, time intervals for each phase, and a transition direction inheritance field. The time intervals for each phase are directly derived from the corresponding moments of the lag-compensation phase boundaries, and the transition direction inheritance field is taken from the transition direction of the postural transition event. In high-risk patients with orthostatic hypotension, sudden drops in blood pressure occur concentrated in the transition period; the phase division identifier labels sampling points within this period as transition period labels. Individualized calibration of the lag-compensation phase boundaries ensures that the boundary moments of the phase division identifier are aligned with the actual physiological response sequence of the patient. When the time intervals of adjacent postural transition event phases overlap, the preparation period of the newer event is prioritized to truncate the previous stable period, ensuring that each sampling point carries only a unique phase type label.

[0023] Step S120: Blood pressure sensor data is sampled in stages by stage division identifier to generate multi-stage blood pressure sampling sequence. Individualized baseline is constructed on the multi-stage blood pressure sampling sequence to generate blood pressure reference curve. Stage comparison mapping is established between blood pressure reference curve and real-time blood pressure sampling value to form stage reference benchmark.

[0024] Specifically, blood pressure sensor data is sampled in stages using stage division markers to generate a multi-stage blood pressure sampling sequence. Each stage time interval of the stage division marker corresponds to approximately 3000 sampling points in the preparation phase, approximately 800 sampling points in the transition phase, and approximately 6000 sampling points in the 100Hz sampling sequence of the blood pressure sensor data, totaling approximately 9800 sampling points, constituting the sample size of the multi-stage blood pressure sampling sequence. The multi-stage blood pressure sampling sequence is composed of three stage sub-sequences, each retaining all stroke-weighted systolic blood pressure values ​​and timestamps within its corresponding time interval. The stage type label sequence of the stage division markers indicates the stage affiliation of each sampling point in the blood pressure sensor data; the multi-stage blood pressure sampling sequence directly inherits this label to retain stage information. In a record of transitioning from a supine to an upright position, the transition period time interval of the stage division markers covers the 8-second window where the baroreceptor is most active in regulation. The sampling values ​​of the blood pressure sensor data within this interval are extracted to form the sub-sequence with the largest fluctuation amplitude in the multi-stage blood pressure sampling sequence. Sampling points with invalid hardware markings in the blood pressure sensor data are simultaneously included in the multi-stage blood pressure sampling sequence during phased sampling. These sampling points carry both an invalid marking and a phase affixation annotation. The multi-stage blood pressure sampling sequence is organized into three sub-sequences indexed by the phase identifier. The start and end timestamps of each sub-sequence are derived from the time interval of each phase division identifier, without pruning or modifying the original blood pressure sensor data.

[0025] In some embodiments, the step of constructing an individualized baseline for the multi-stage blood pressure sampling sequence to generate a blood pressure baseline curve includes: performing resting-state stability screening on the multi-stage blood pressure sampling sequence to generate a stability qualification marker; performing outlier detection on the multi-stage blood pressure sampling sequence based on the stability qualification marker to identify outlier sampling points; performing exclusion correction on the outlier sampling points to generate a corrected sampling sequence; and using the corrected sampling sequence to perform multi-stage segmented fitting to generate a blood pressure baseline curve.

[0026] Resting-state stability screening was performed on multi-stage blood pressure sampling sequences to generate a stability qualification mark. Blood pressure values ​​in the preparation phase subsequence of the multi-stage blood pressure sampling sequence exhibit low-amplitude fluctuations, with a systolic blood pressure standard deviation typically below 5 mmHg, while the systolic blood pressure standard deviation in the transition phase subsequence can reach above 20 mmHg. The difference in stability between the two subsequences forms the physiological basis for setting the screening threshold. Resting-state stability screening divided each phase subsequence of the multi-stage blood pressure sampling sequence into short time windows of 30 seconds. Sampling points carrying invalid markers within the short time window were not included in the standard deviation calculation; the systolic blood pressure standard deviation of the remaining valid sampling points was used as the stability score for that window. When the short-time window stability score of the multi-stage blood pressure sampling sequence was below the threshold of 8 mmHg, the stability qualification mark indicated that the sampling points within that window were qualified; when the score exceeded the threshold, the stability qualification mark indicated that the corresponding sampling points were unqualified. When the patient is lying supine and at rest, systolic blood pressure drifts slowly between 118 and 122 mmHg, with a standard deviation of approximately 2 mmHg within a 30-second window. The stability score is far below the threshold, and all sampling points are marked as acceptable. After standing up, during the transition period, systolic blood pressure drops sharply from over 120 mmHg and then gradually rises. Within the same time window, the standard deviation exceeds 15 mmHg, and the stability score exceeds the threshold. The corresponding sampling points are marked as unacceptable. A large number of sampling points in the transition period subsequence of the sampling sequence are marked as unacceptable by the stability acceptance marker, with the proportion of unacceptable sampling points typically exceeding 70%. The proportion of unacceptable sampling points in the preparation and stable periods is typically less than 20%. The stability acceptance marker performs binary labeling on all sampling points in the multi-stage blood pressure sampling sequence, with acceptable sampling points mainly concentrated in the preparation and stable periods.

[0027] Outlier detection is performed on multi-stage blood pressure sampling sequences based on stability qualification markers. The qualification status of the stability qualification markers filters out qualified sampling points in the multi-stage blood pressure sampling sequences as the reference population for outlier detection. Sampling points marked as unqualified by the stability qualification markers are excluded from the reference population, ensuring that the statistical baseline is not affected by fluctuation periods. Outlier detection is performed independently within each stage subsequence of the multi-stage blood pressure sampling sequence. The mean and standard deviation of each stage subsequence are established using the blood pressure values ​​of qualified sampling points with stability qualification markers. The statistics within each stage are independent to accommodate different physiological states at each stage. When the blood pressure value of a sampling point in the multi-stage blood pressure sampling sequence differs from the mean of its stage by more than three times the standard deviation, that sampling point is identified as an outlier. The three-times-standard-deviation threshold corresponds to an extreme deviation with a probability of less than 0.3% under a normal distribution. In patients with orthostatic hypotension, systolic blood pressure can drop sharply to below 60 mmHg during the transition period. In multi-stage blood pressure sampling sequences, the density of sampling points exceeding the statistical threshold in this stage is higher than in the preparation stage. Outlier sampling points identified during the transition period typically account for more than 60% of all outlier sampling points. Unqualified sampling points from the stability-qualified markers are separately incorporated into the outlier sampling point set during outlier detection in the multi-stage blood pressure sampling sequence. Sampling points with invalid hardware labels in the blood pressure sensor data are simultaneously incorporated into the outlier sampling point set. Outlier sampling points from all three sources undergo the same processing in subsequent correction.

[0028] Outlier sampling points are excluded and corrected to generate a corrected sampling sequence. The exclusion correction operates on a multi-stage blood pressure sampling sequence, replacing all outlier blood pressure values ​​with invalid placeholders, indicating that the sampled value at that moment has been excluded. The proportion of outliers in each stage subsequence determines the effective sampling point density of the corrected sampling sequence. When there are many outliers during the transition period, the spacing between effective points in the corrected sampling sequence for that stage increases accordingly. After outliers are excluded, if consecutive invalid placeholders exist between adjacent effective sampling points, the corrected sampling sequence is completed using linear interpolation within that interval. The boundary anchor points required for interpolation are provided by the adjacent non-outlier sampling points. For example, a wrist movement at the moment the patient stands up causes a continuous 0.3-second sampling anomaly; after exclusion, the anchor points on both sides represent the stable sampling values ​​before and after the jump. Linear interpolation makes the corrected sampling sequence show a continuous downward trend in this segment rather than abnormal jagged edges. When outlier sampling points appear consecutively for a duration exceeding 5 seconds, the corrected sampling sequence is marked with a large missing region within that interval. No linear interpolation is performed; instead, invalid placeholders are directly retained. The range information of the large missing region is stored as a data quality annotation in the corrected sampling sequence. The corrected sampling sequence is organized into three parts: a timestamp sequence, a corrected blood pressure value sequence, and data quality annotations. Valid point values ​​in the corrected blood pressure value sequence originate from the original blood pressure sensor data, while invalid placeholders are derived from outlier exclusion markings or the retention of large missing regions.

[0029] A blood pressure baseline curve is generated by multi-stage segmented fitting using the corrected sampled sequence. In the preparation phase of the corrected sampled sequence, the effective sampling points are densely distributed, and the fluctuation range of blood pressure values ​​is small. The fitting segment in the preparation phase uses a quadratic polynomial to obtain a smooth baseline, with the fitting formula being B(t) = a0 + a1t + a2t², where B(t) is the estimated baseline blood pressure at time t (in mmHg), a0, a1, and a2 are the fitting coefficients, and t is the relative time (in seconds) with the start time of the phase as zero. The root mean square of the fitting residuals is typically less than 2 mmHg. Multi-stage segmented fitting establishes independent fitting models for the three phase sub-sequences of the corrected sampled sequence. A quadratic polynomial is used for the preparation and stabilization phases, and a cubic polynomial is used for the transition phase to accommodate the valley-shaped characteristic of an initial decrease followed by a rise. The three-segment fitting models are subject to dual constraints at the phase boundaries: continuity of function values ​​and continuity of the first derivative. The constraint equations are as follows: and Where t_b is the boundary time between adjacent stages, and B_i and B_{i+1} are the fitted models on both sides of the boundary, respectively. and The first derivative is used as the basis for the dual constraints to ensure that the blood pressure baseline curve has no abrupt changes at the stage boundaries and that the rate of change transitions smoothly. The range information of large missing regions in the corrected sampling sequence is used as the exclusion interval for the fitting constraints. The fitting algorithm skips the sampling points in this interval and fills it with extrapolated effective anchor points on both sides. The corresponding positions of the large missing regions are marked with extrapolated estimates in the blood pressure baseline curve. The blood pressure baseline curve typically exhibits a valley-shaped characteristic of first decreasing and then increasing during the transition phase. The delay between the trough and the trigger time of the postural change event reflects the response rate of the individual baroreceptor. After piecewise fitting of the effective sampling points of the corrected sampling sequence, the blood pressure baseline curve is stored in two parts: a timestamp sequence and a baseline blood pressure value sequence. The baseline blood pressure value sequence is calculated and output by the piecewise fitting model of the corrected sampling sequence at a 100Hz timestamp resolution.

[0030] In some embodiments, establishing a stage reference benchmark by comparing the blood pressure baseline curve with the real-time blood pressure sampled values ​​includes: assessing the blood pressure variability of the blood pressure baseline curve and the real-time blood pressure sampled values ​​to extract the variability level corresponding to each stage; configuring stage weight coefficients according to the variability level corresponding to each stage; performing a weighted comparison on the blood pressure baseline curve and the real-time blood pressure sampled values ​​based on the stage weight coefficients to generate a weighted deviation component; and using the weighted deviation component to construct a stage weight mapping table to generate a stage reference benchmark.

[0031] Blood pressure variability was assessed using a baseline blood pressure curve and real-time blood pressure samples to extract the variability level for each stage. Real-time blood pressure samples are the continuous output segment of blood pressure sensor data following the current positional change event. This sequence shares the same 100Hz timestamp resolution as the baseline blood pressure curve and participates in variability assessment through point-by-point comparison. The coefficient of variation (CV) was extracted from both the baseline blood pressure curve and real-time blood pressure samples for each stage. The CV is defined as the ratio of the standard deviation to the mean within a stage, calculated as CV = σ / μ × 100%, where CV is the CV (in %), σ is the standard deviation of blood pressure within a stage (in mmHg), and μ is the mean blood pressure within a stage (in mmHg). A larger CV indicates a higher relative amplitude of blood pressure fluctuation within that stage. During the patient's preparation phase, the mean blood pressure baseline curve at supine rest has approximately 120 mmHg and a standard deviation of approximately 2 mmHg, with a coefficient of variation of approximately 1.7%, which is considered low-grade. During the transition phase, the standard deviation of real-time blood pressure samples can reach 18 mmHg, and the coefficient of variation exceeds 15%, which is considered high-grade. The difference in grades between the two phases directly drives the differentiated configuration of phased weighting coefficients. The variability grades corresponding to each phase are divided into three levels: low, medium, and high, based on the range of coefficient of variation values. A coefficient of variation below 5% is considered low-grade, 5% to 15% is medium-grade, and above 15% is high-grade. This grading rule applies uniformly to all phases of both the blood pressure baseline curve and real-time blood pressure samples. The variability grades corresponding to each phase cover all three phases of both the blood pressure baseline curve and real-time blood pressure samples, totaling six grade values, stored in a structure containing a phase identifier, the blood pressure baseline variability grade, and the real-time blood pressure sample variability grade.

[0032] The weighting coefficients for each stage are configured based on the variability level corresponding to each stage. Lower variability levels correspond to higher baseline weighting coefficients, while higher variability levels correspond to lower baseline weighting coefficients. This reverse mapping logic ensures that stages with stable blood pressure fluctuations receive higher confidence weights in the comparison calculation. The weighting coefficients for each stage use the variability level as the mapping input. The mapping rule from level to weight is: low level = 0.9, medium level = 0.7, and high level = 0.5. For example, a low variability level in the preparation phase corresponds to a weighting coefficient of 0.9, a high variability level in the transition phase corresponds to a weighting coefficient of 0.5, and a medium variability level in the stable phase corresponds to a weighting coefficient of 0.7. In each stage, the contribution of the blood pressure baseline variability level and the real-time blood pressure sample level to the stage weighting coefficient is unequal. The contribution coefficient of the blood pressure baseline variability level is set to 0.6, and the contribution coefficient of the real-time blood pressure sample level is set to 0.4. The final stage weighting coefficient is a weighted sum of the two, with the formula W = 0.6 × f(L_base) + 0.4 × f(L_real), where W is the final stage weighting coefficient, L_base is the variability level of the blood pressure baseline, L_real is the variability level of the real-time blood pressure sample, and f() is the mapping function from level to weight (low → 0.9, medium → 0.7, high → 0.5). The stage weighting coefficients are stored in an array with the stage identifier as the key, and each of the three stages corresponds to a weight value. The mapping relationship between the variability level corresponding to each stage and the stage weighting coefficient remains fixed. When the variability level corresponding to each stage is updated, the stage weighting coefficient is reconfigured according to the new variability value.

[0033] A weighted bias component is generated by performing a weighted comparison between the blood pressure baseline curve and real-time blood pressure samples based on phased weighting coefficients. The weighted comparison calculates the original bias by subtracting the baseline estimate of the blood pressure baseline curve from the stroke-weighted systolic pressure of the real-time blood pressure sample. The original bias is then multiplied by the weight value of the corresponding phase in the phased weighting coefficients to obtain the weighted bias value, calculated as D_w(t) = W_i × (P(t) − B(t)), where D_w(t) is the weighted bias value at time t (in millimeters of mercury), P(t) is the stroke-weighted systolic pressure of the real-time blood pressure sample at time t, B(t) is the baseline estimate of the blood pressure baseline curve at time t, and W_i is the phased weighting coefficient corresponding to phase i at time t. The timestamp sequence of the blood pressure baseline curve and the real-time blood pressure samples are aligned point-by-point based on timestamps; pairs with timestamp differences within 2 milliseconds are considered successfully aligned. The phased weighting coefficients assign lower weight values ​​to the transition period than to the preparation and stabilization periods, reducing the magnitude of the weighted bias component in this phase and mitigating the excessive impact of drastic fluctuations on the overall bias assessment. In patients with orthostatic hypotension, the weighted bias component during the transition period exhibits a characteristic negative value segment. The amplitude and duration of this negative segment are key features for identifying orthostatic hypotension. The weighted bias component is stored in two parts: a timestamp sequence and a weighted bias value sequence.

[0034] A stage weight mapping table is constructed using weighted deviation components to generate a stage reference benchmark. The time interval information of the stage reference benchmark comes from the stage division identifier. After aligning the timestamp sequence of the weighted deviation components with the time intervals of each stage of the stage division identifier, the stage weight mapping table assigns the weighted deviation value sequence of the weighted deviation components to the corresponding stage storage cells using the stage identifier as an index. The stage weight mapping table extracts statistics for each of the three stage segments of the weighted deviation components. The statistics include the mean and standard deviation of the weighted deviation components in each stage. These two types of statistics constitute the core evaluation basis of the stage reference benchmark. In patients with orthostatic hypotension, the mean of the weighted deviation component is close to zero when lying supine during the preparation phase, reflecting that the real-time blood pressure value at baseline deviates very little from the baseline curve. After getting up and entering the transition phase, the mean of the weighted deviation component is significantly negative, quantifying the weighted magnitude of the overall decrease in blood pressure relative to the baseline curve. The difference between the means of the two stages is directly presented as a cross-stage deviation comparison dimension in the stage weight mapping table. The ratio of the standard deviations of the weighted deviation components across each stage is directly presented in the stage weight mapping table. The standard deviation during the transition period is typically 3 to 5 times that of the preparation period. This ratio reflects the amplification factor of blood pressure fluctuations between stages caused by positional changes. The stage reference benchmark uses the stage weight mapping table as its data source and includes the deviation statistics for each stage and the time intervals for each stage. The deviation statistics for each stage of the stage reference benchmark are derived from the calculation results of the weighted deviation components for each stage segment, and the time intervals for each stage are derived from the stage division identifiers.

[0035] Step S130: Calculate the deviation of the real-time blood pressure sampling value point by point according to the stage reference benchmark to generate a blood pressure deviation sequence. Detect the continuous downward trend and recovery characteristics after the decline in the blood pressure deviation sequence to generate rate constraint parameters. Apply rate constraint to the blood pressure deviation sequence based on the rate constraint parameters to form a dynamic blood pressure change trajectory.

[0036] A blood pressure deviation sequence is generated by calculating the deviation of real-time blood pressure samples point by point based on a phase reference baseline. The mean deviation of the phase reference baseline during the preparation period is usually within ±3 mmHg, while the mean deviation during the transition period can reach below -15 mmHg. The difference between the means of the two phases establishes a reference level for phase-specific deviation calculation. The stroke-weighted systolic blood pressure value of the real-time blood pressure sample within each phase time interval is subtracted point by point from the mean deviation of the corresponding phase in the phase reference baseline. The deviation calculation formula is E(t) = P(t) − μ_i, where E(t) is the deviation value at time t (in mmHg), P(t) is the stroke-weighted systolic blood pressure of the real-time blood pressure sample at time t, and μ_i is the mean deviation of phase i to which time t belongs. The mean removal operation eliminates the interference of the systematic deviation level of each phase on the identification of individual fluctuations in the blood pressure deviation sequence. When the patient is supine, the blood pressure deviation sequence oscillates slightly around zero during the preparation phase, then curves downward to a negative trough during the transition phase after standing up, and slowly rises again. During the stable phase, it gradually converges back to zero. These three segments reflect the complete temporal process of blood pressure regulation induced by postural changes. The blood pressure deviation sequence is stored in three parts: a timestamp sequence, a deviation value sequence, and a stage assignment. Each value in the deviation value sequence is calculated from the corresponding sampling point of the real-time blood pressure sample and the mean value of the corresponding stage reference baseline. The stage assignment is derived from the time intervals of each stage reference baseline.

[0037] In some embodiments, the step of detecting a continuous downward trend and generating rate constraint parameters for the blood pressure deviation sequence and its recovery characteristics after the decline includes: performing time-period analysis on the blood pressure deviation sequence to identify continuous downward deviation segments; performing valley location and recovery inflection point identification on the continuous downward deviation segments to generate a decline-recovery time window; extracting rate features from the continuous downward deviation segments based on the decline-recovery time window to generate segmented rate thresholds; and determining rate constraint parameters based on the segmented rate thresholds.

[0038] Time-phase analysis was performed on blood pressure deviation sequences to identify continuously decreasing deviation segments. In the deviation value subsequence covered by the transition phase of the blood pressure deviation sequence, patients with orthostatic hypotension typically exhibit a monotonically decreasing pattern over 15 to 30 consecutive sampling points, with a decrease of up to -20 mmHg. This continuous decreasing pattern is the core objective for identifying continuously decreasing deviation segments. Time-phase analysis used the deviation value sequence of the blood pressure deviation sequence as input, and detected time periods that continuously met the deviation value decreasing condition frame by frame using a 0.5-second sliding window. The decreasing condition was defined as the deviation value remaining monotonically decreasing within 0.5 seconds and the cumulative decrease exceeding 5 mmHg. The phase assignment of the blood pressure deviation sequence constrained the identification range of continuously decreasing deviation segments. The start time of a continuously decreasing deviation segment must fall within the transition phase or within 1 second of the transition phase's leading edge to avoid misinterpreting normal baseline fluctuations during the preparation phase as a decreasing segment. The identification results of continuously decreasing deviation segments were recorded as the time interval of the decreasing segment, which includes the start and end times of the decrease. The end time of the decrease was defined as the time of the last sampling point that met the decreasing condition. When the interval between two adjacent candidate drops in the same postural change event within a blood pressure deviation sequence is less than 2 seconds, the time interval analysis performs a merging process. After merging, the time interval of the drop segment is redefined based on the earliest start time and the latest end time of the drop. For example, after a patient stood up, the deviation value dropped continuously from -2 mmHg to -22 mmHg in approximately 3 seconds. The time interval of the continuous drop deviation segment completely captured this 4-second sudden drop. The continuous drop deviation segment includes the time interval of the drop segment and the extreme value of the drop segment deviation. The extreme value of the drop segment deviation is the minimum value of the deviation value sequence within that time interval. Both are directly determined by the time interval analysis results of the blood pressure deviation sequence.

[0039] For a continuous decline in deviation, trough location and recovery inflection point identification are performed to generate a decline-recovery time window. Trough location and recovery inflection point identification use the deviation value sequence of the blood pressure deviation sequence as the search object, and are performed within the time range defined by the continuous decline in deviation. The extreme value of the decline segment of the continuous decline in deviation gives the lowest deviation point during the decline process. Trough location precisely searches for sampling points in the deviation value sequence of the blood pressure deviation sequence that coincide with the extreme value of the decline segment. Trough location uses the time interval of the decline segment of the continuous decline in deviation as the search boundary. Within this boundary, the timestamp of the sampling point containing the minimum deviation value sequence is found to determine the trough time. Sampling points outside the boundary are not included in trough location. Recovery inflection point identification searches for the turning point in the blood pressure deviation sequence after the trough time where the deviation value changes from a monotonically decreasing trend to a continuous positive increase. The turning point determination condition is that the linear regression slope of the deviation value within the 0.5-second window after the trough is positive, and the deviation value at the end of the window is at least 1 mmHg higher than the deviation value at the beginning of the window. After a patient's deviation value plateaued for approximately 1.5 seconds, it began to steadily rise. The recovery inflection point identification skipped the plateau phase, recording the first sampling point that met the condition of continuous positive increase as the recovery inflection point. The descent-recovery time window starts at the beginning of the continuous descent deviation segment and ends at the recovery inflection point. The trough value divides the entire window into a descent sub-segment and a recovery sub-segment. The duration of the descent sub-segment is the difference between the trough value and the beginning of the descent, and the duration of the recovery sub-segment is the difference between the recovery inflection point and the trough value. The descent-recovery time window includes both the trough value and the recovery inflection point, both determined by the deviation value sequence analysis results of the continuous descent deviation segment.

[0040] Rate feature extraction is performed on continuous downward deviation segments based on the descent-recovery time window to generate segmented rate thresholds. Rate feature extraction is based on the point-by-point rate obtained by dividing the difference in deviation values ​​between adjacent sampling points within each sub-segment of the blood pressure deviation sequence by the sampling interval. The trough moment of the descent-recovery time window divides the continuous downward deviation segment into a descent sub-segment and a recovery sub-segment, and rate feature extraction is performed independently for each sub-segment. The point-by-point difference of the descent sub-segment constitutes the descent rate sample set, and the 75th percentile of the sample set is used as the representative value of the descent rate. The duration of the descent sub-segment within the descent-recovery time window determines the size of the sample set, and this representative value serves as the descent rate threshold for the segmented rate threshold. The point-by-point difference of the recovery sub-segment forms the recovery rate sample set, and the 25th percentile of the sample set is used as the representative value of the recovery rate. A bias towards the slower end corresponds to a slower physiological recovery state, and this representative value serves as the recovery rate threshold for the segmented rate threshold. When multiple candidate descent events exist within a continuous descent deviation segment, rate feature extraction primarily focuses on the main descent-recovery time window with the largest amplitude. Rate samples from other candidate events are merged into the sample set. The sample sets from multiple consecutive descent deviation segments are then combined to uniformly determine the segmented rate threshold. The segmented rate threshold includes a descent rate threshold and a recovery rate threshold, which are determined by the rate representative values ​​of the two sub-segments, respectively.

[0041] Rate constraint parameters are determined based on segmented rate thresholds. The construction of the constraint window depends on the time interval of the continuous descent deviation segment and the recovery inflection point of the descent-recovery time window. The segmented rate threshold determines the constraint strength based on this. The upper limit of the descent rate of the rate constraint parameter is determined by multiplying the descent rate threshold of the segmented rate threshold by a tolerance coefficient of 1.2, and the lower limit of the recovery rate is determined by multiplying the recovery rate threshold of the segmented rate threshold by a tolerance coefficient of 0.8. The tolerance coefficient introduces an appropriate margin to prevent the critical rate sampling point from being erroneously truncated, and the tolerance coefficient prevents excessive smoothing from masking the subtle recovery signal at the beginning of the recovery phase. The constraint window identifier of the rate constraint parameter is obtained by extending the time interval of the continuous descent deviation segment by 1 second before the start of the descent and 1 second after the recovery inflection point. The entire process of sudden drop and recovery from the trough in patients with orthostatic hypotension falls within the constraint window. Sampling points at the end of the preparation phase and the stable phase outside the window are not affected by rate constraints and retain the original deviation values ​​for baseline and recovery status assessment. The rate constraint parameters include the upper limit of the descent rate, the lower limit of the recovery rate, the constraint window identifier, and the valley time. The first three are established based on individualized values ​​of the segmented rate thresholds. The valley time is inherited from the descent-recovery time window and is used to determine the switching boundary between the descent segment and the recovery segment when the constraint is executed, so as to ensure that the constraint direction matches the patient's actual physiological rate.

[0042] A dynamic trajectory of blood pressure changes is formed by applying rate constraints to the blood pressure deviation sequence based on rate constraint parameters. The constraint window identifier of the rate constraint parameters covers the continuously decreasing deviation segment and its preceding and following buffers. No rate constraints are applied to sampling points outside the window, and the deviation values ​​of the blood pressure deviation sequence within the window are constrained point by point. Rate constraints are applied point-by-point in a sliding manner on the deviation value sequence of the blood pressure deviation sequence. The constraint rule for the descent phase is E'(t) = max(E(t), E'(t-1) - R_down × Δt), and the constraint rule for the recovery phase is E'(t) = min(E(t), E'(t-1) + R_rec × Δt), where E'(t) is the deviation value after constraint at time t (in mmHg), E(t) is the original deviation value at time t (in mmHg), E'(t-1) is the deviation value after constraint at the previous sampling point (in mmHg), R_down is the upper limit of the descent rate of the rate constraint parameter, R_rec is the lower limit of the recovery rate of the rate constraint parameter, and Δt is the sampling interval of 0.01 seconds. The two constraint rules are assigned based on the valley value of the rate constraint parameter to determine the sub-segment, preventing directional mis-constraints. After rate constraint, the blood pressure deviation sequence retains the sudden drop trend and eliminates abnormal jumps caused by sampling noise. Together with the phase assignment of the blood pressure deviation sequence, it constitutes the dynamic change trajectory of blood pressure. The blood pressure dynamic change trajectory is organized into three parts: trajectory timestamp sequence, trajectory deviation value sequence, and stage assignment. The trajectory deviation value sequence is output by constraining the blood pressure deviation sequence with rate constraint parameters, and the stage assignment of the blood pressure dynamic change trajectory is directly inherited from the blood pressure deviation sequence.

[0043] Step S140: Separate the physiological rhythm fluctuations of the blood pressure dynamic change trajectory to determine the abnormal fluctuation range, and perform risk assessment on the blood pressure dynamic change trajectory based on the abnormal fluctuation range to form a stage risk assessment map.

[0044] In some embodiments, the step of separating the physiological rhythm fluctuations of the blood pressure dynamic change trajectory to determine the abnormal fluctuation range includes: performing local mean deviation analysis on the blood pressure dynamic change trajectory to generate a fluctuation residual sequence; performing individualized respiratory cycle estimation on the fluctuation residual sequence to generate an adaptive rhythm template; performing template matching separation on the fluctuation residual sequence based on the adaptive rhythm template to extract non-rhythmic fluctuation components; and determining the abnormal fluctuation range based on the amplitude and significance score of the non-rhythmic fluctuation components.

[0045] Local mean deviation analysis (LMDE) is performed on the dynamic blood pressure trajectory to generate a fluctuation residual sequence. The trajectory deviation value sequence of the dynamic blood pressure trajectory is used to calculate the sliding local mean with a window length of 30 seconds and a step size of 0.1 seconds. The 30-second window covers more than twice the longest period of the Mayer wave (12 seconds), ensuring that the local mean only captures the slow trend without eliminating respiratory rhythm and Mayer wave rhythm components. LMD subtracts the sliding mean at the corresponding time from the value of each sampling point in the trajectory deviation value sequence of the dynamic blood pressure trajectory. The residual calculation formula is R(t) = X(t) − M(t), where R(t) is the residual value at time t (in mmHg), X(t) is the value of the trajectory deviation value sequence of the dynamic blood pressure trajectory at time t, and M(t) is the mean of the trajectory deviation values ​​within the 30-second sliding window centered at t. The residual removes the slow-changing trend component from the trajectory deviation value sequence. The stage assignment of the blood pressure dynamic trajectory is used to independently initialize the moving average window within each stage. When sampling points from different stages are mixed into the window at the stage boundary, the window is truncated at the stage boundary and reinitialized to prevent cross-stage aliasing from introducing trend bias. During the preparation period, the trajectory deviation value sequence drifts slowly by about 2 mmHg, and the residual values ​​of the fluctuation residual sequence are concentrated within ±1.5 mmHg. During the transition period, rhythmic components and abnormal components overlap, and the residual amplitude of the fluctuation residual sequence is significantly larger than that during the preparation period. The fluctuation residual sequence is stored in two parts: a residual value sequence and a residual timestamp sequence. The residual value sequence is calculated by subtracting the moving average from the trajectory deviation value sequence of the blood pressure dynamic trajectory point by point, and the residual timestamp sequence is consistent with the trajectory timestamp sequence of the blood pressure dynamic trajectory.

[0046] For example, the step of generating an adaptive rhythm template by performing individualized respiratory cycle estimation on the fluctuating residual sequence includes: performing periodic analysis on the fluctuating residual sequence to generate a residual cycle distribution map; extracting the main respiratory wave cycle and the main Mayer wave cycle from the residual cycle distribution map to determine the dual-band physiological rhythm reference cycle; performing adaptive cycle tracking on the fluctuating residual sequence based on the dual-band physiological rhythm reference cycle to generate a dynamic rhythm waveform; and constructing an adaptive rhythm template based on the dynamic rhythm waveform.

[0047] Periodic analysis is performed on the fluctuation residual sequence to generate a residual period distribution map. The residual value sequence of the fluctuation residual sequence is used to calculate the power spectral density through Fast Fourier Transform (FFT). The transform window length is adaptively set according to the data duration of each stage. During the stable period, a 60-second window is used, corresponding to 6000 sampling points at a sampling rate of 100Hz. When the data duration of the preparation and transition periods is less than 60 seconds, the actual duration is used as the window and zeros are padded to the nearest power of 2 number of sampling points to meet the FFT calculation requirements. The frequency resolution is determined by the reciprocal of the actual window duration. The Mayer wave main period extraction is fixed on the power spectral density of the preparation or stable period as the data source. The data duration of the transition period is insufficient to meet the resolution requirements of the Mayer wave band and is not included in the Mayer wave main frequency extraction. The residual sequence in the preparation phase of the oscillating residual sequence typically exhibits a clear power spectral peak around 0.25 Hz, corresponding to a resting respiratory rate of 15 breaths per minute. The amplitude and width of the power spectral peak reflect the stability of the individual's respiratory rhythm. After standing up, breathing becomes deeper and faster due to stress, causing the power spectral peak to shift towards higher frequencies and broaden. The residual period distribution map visually records the dynamic changes in respiratory pattern with body position. In the transition phase, the power spectral density of the oscillating residual sequence typically increases in amplitude within the 0.1 to 0.15 Hz frequency band compared to the preparation phase, corresponding to the Mayer wave enhancement phenomenon triggered by body position change. This frequency band change is a key basis for extracting the Mayer wave main period. The power spectral density of the three phases of the oscillating residual sequence is calculated independently. Independent calculation of each phase avoids spectral aliasing between different physiological states. The power spectral density of each phase is stored at the corresponding phase position in the residual period distribution map. The residual period distribution map is organized and stored according to the frequency-power spectral density correspondence of each phase. The peak positions of the power spectral density array in each target frequency band are used for main period extraction.

[0048] The respiratory wave principal period and Mayer wave principal period are extracted from the residual period distribution map to determine the baseline cycle of the dual-frequency physiological rhythm. The peak frequency of the power spectral density array in the 0.2–0.4 Hz band of the residual period distribution map corresponds to the respiratory wave principal frequency. The principal period of the respiratory wave is determined by the reciprocal of this peak frequency, which is typically between 2.5 and 5 seconds. The Mayer wave principal frequency in the residual period distribution map is detected in the 0.08–0.12 Hz band. The peak position of the power spectral density array in this band is the Mayer wave principal frequency, and the Mayer wave principal period is the reciprocal of this frequency, typically between 8 and 12 seconds. The respiratory wave principal period of the baseline cycle of the dual-frequency physiological rhythm is determined by the reciprocal of the peak value in the 0.2–0.4 Hz band of the residual period distribution map, and the Mayer wave principal period is determined by the reciprocal of the peak value in the 0.08–0.12 Hz band of the residual period distribution map. These two principal periods together describe the dual-rhythm cycle characteristics of an individual in a given recording. When the power spectral density array of the residual period distribution plot does not have a clear peak in the target frequency band, the dual-band physiological rhythm baseline period is filled with population reference values. The default main period of the respiratory wave is 4 seconds, and the default main period of the Mayer wave is 10 seconds, with the data source labeled as population reference values. Both main periods in the dual-band physiological rhythm baseline period are directly derived from the power spectral analysis results of the residual period distribution plot, with a period accuracy corresponding to the reciprocal accuracy of the frequency resolution of 0.017 Hz.

[0049] Based on the dual-band physiological rhythm baseline period, adaptive periodic tracking is performed on the fluctuation residual sequence to generate dynamic rhythm waveforms. The main respiratory wave period and the main Mayer wave period of the dual-band physiological rhythm baseline period are used as the initial frequency parameters for adaptive periodic tracking. The tracking algorithm sets a frequency search range of ±20% centered on each main period, and continuously optimizes the period estimation within the search range to adapt to the slow drift of individual rhythms. The residual value sequence of the fluctuation residual sequence is used as the observation input in adaptive periodic tracking. The tracking algorithm extracts the rhythmic components that match the dual-band physiological rhythm baseline period from the time domain structure of this sequence. The extraction process aims to minimize the tracking residual. The main respiratory wave period of the dual-band physiological rhythm baseline period corresponds to the high-frequency channel of the tracking algorithm, and the main Mayer wave period corresponds to the low-frequency channel. The tracking results of the two channels are superimposed in the time domain to form a dual-rhythm synthetic output. During a patient's transition period, the respiratory rate increased from 0.25 Hz at rest to 0.35 Hz. Adaptive cycle tracking was initialized based on the main cycle of the respiratory wave of the dual-band physiological rhythm reference cycle. The cycle estimate was continuously updated within the tracking window. The period of the high-frequency channel of the dynamic rhythm waveform gradually shortened from 4 seconds to 2.9 seconds during this period. The dynamic rhythm waveform is stored in two parts: a waveform value sequence and a corresponding timestamp sequence. The waveform value sequence of the dynamic rhythm waveform is synthesized by superimposing the dual-channel tracking results of the dual-band physiological rhythm reference cycle. The timestamp sequence is aligned with the residual timestamp sequence of the fluctuation residual sequence.

[0050] An adaptive rhythm template is constructed based on dynamic rhythm waveforms. The waveform value sequence of the dynamic rhythm waveform includes the superposition contributions of the respiratory channel and the Mayer wave channel. The adaptive rhythm template extracts the peak and trough amplitudes of the dynamic rhythm waveform cycle by cycle, scales the waveform of each cycle to a unit amplitude, and performs phase alignment to form a standard shape waveform with normalized amplitude. The timestamp sequence of the dynamic rhythm waveform defines the temporal coverage of the adaptive rhythm template. The temporal resolution of the template is consistent with the sampling rate of the waveform value sequence to ensure that there is no interpolation error when the template is matched point by point with the fluctuation residual sequence. The amplitude parameter of the adaptive rhythm template records the original peak and trough amplitudes of each complete cycle in the dynamic rhythm waveform and is updated smoothly by adjacent cycles. When matching the template, the normalized standard waveform is multiplied by the amplitude parameter of the corresponding time period to restore the actual rhythm amplitude for subtraction calculation. After the patient gets up, breathing becomes deeper, and the peak and trough amplitudes during the transition period are significantly greater than those during the preparation period. The amplitude parameter of the adaptive rhythm template is adjusted up synchronously. If the fixed amplitude during the preparation period is used, the residual rhythm fluctuations will mix in non-rhythmic components, interfering with the identification of abnormal fluctuations. During the transition period, the rhythm amplitude of the dynamic rhythm waveform is usually higher than that during the preparation period. The amplitude parameter of the adaptive rhythm template is adjusted accordingly during the transition period, improving the rhythm separation accuracy in this stage. The adaptive rhythm template consists of a template waveform sequence and a template overlay timestamp sequence. The template waveform sequence is obtained by normalizing the waveform value sequence of the dynamic rhythm waveform and adaptively adjusting the amplitude. The template overlay timestamp sequence is perfectly aligned with the corresponding timestamp sequence of the dynamic rhythm waveform.

[0051] Template matching separation is performed on the fluctuation residual sequence based on an adaptive rhythm template to extract non-rhythmic fluctuation components. The template waveform sequence of the adaptive rhythm template and the residual value sequence of the fluctuation residual sequence are aligned point-by-point within the range marked by the template coverage timestamp sequence. After alignment, point-by-point subtraction is performed, and the difference constitutes an initial non-rhythmic candidate sequence. Template matching separation dynamically scales the template waveform sequence using the amplitude parameter of the adaptive rhythm template. The scaling factor is determined by minimizing the mean square error between the residual value sequence and the template within each rhythm cycle. Local optimal scaling ensures that the rhythm component is sufficiently subtracted without introducing over-subtraction. The portion of the fluctuation residual sequence that the adaptive rhythm template cannot completely match forms the matching residual. The matching residual manifests as irregular fluctuations outside the rhythm frequency in the time domain, constituting the main source of non-rhythmic fluctuation components. For fluctuation residual sequence sampling points outside the range covered by the template coverage timestamp sequence of the adaptive rhythm template, template matching separation directly copies these residual values ​​as non-rhythmic fluctuation components without performing rhythm subtraction. During a patient's transition period, a sudden drop in blood pressure was observed. The residual sequence of the fluctuation residual sequence was superimposed with a respiratory rhythm fluctuation of approximately 3 mmHg. After subtracting this rhythm component from the adaptive rhythm template, the non-rhythmic fluctuation component clearly showed a pure abnormal decrease in amplitude of approximately -12 mmHg. The non-rhythmic fluctuation component is stored in two parts: a non-rhythmic amplitude sequence and a component timestamp sequence. The non-rhythmic amplitude sequence of the non-rhythmic fluctuation component is calculated by subtracting the scaled result of the template waveform sequence of the adaptive rhythm template from the residual sequence of the fluctuation residual sequence. The component timestamp sequence is consistent with the residual timestamp sequence of the fluctuation residual sequence.

[0052] Abnormal fluctuation ranges are determined based on the amplitude and significance score of the non-rhythmic fluctuation component. The non-rhythmic amplitude sequence of the non-rhythmic fluctuation component oscillates slightly around zero during the preparation period, with a mean absolute value below 2 mmHg. Amplitude segments exceeding three times this benchmark are considered candidate abnormal regions. The significance score considers both the absolute amplitude and duration. A negative segment with a sudden drop in blood pressure lasting several seconds after the patient stands up, exceeding the significance score threshold, is identified as an abnormal fluctuation range. Single-point spikes caused by momentary sensor interference are effectively excluded due to their extremely short duration. Continuous periods with absolute values ​​of non-rhythmic amplitude sequences exceeding 6 mmHg are marked as high-amplitude candidate segments. The significance score is calculated using the formula S = T_seg × A_avg, where S is the significance score (in mmHg·s), T_seg is the duration of the candidate segment (in seconds), and A_avg is the mean of the absolute values ​​of the non-rhythmic amplitude sequences within that candidate segment (in mmHg). Candidate segments with S exceeding 30 mmHg·s are directly identified as anomalous fluctuation intervals. Candidate segments with S between 15 and 30 mmHg·s are marked as secondary threshold candidate segments. When the sum of the significance scores of secondary threshold candidate segments within the same period exceeds 30 mmHg·s, ​​these secondary threshold candidate segments are merged to determine the anomalous fluctuation interval. Component timestamp sequences are used to extract the start and end times of each candidate segment. Typically, 1 to 3 anomalous fluctuation intervals are identified within the transition period for non-rhythmic fluctuation components, with the core anomalous interval having the highest significance score. The abnormal fluctuation range includes the start and end times of the range and a significance score. The start and end times of the range are determined by the component timestamp sequence of the non-rhythmic fluctuation component, and the significance score is calculated by the non-rhythmic amplitude sequence of the non-rhythmic fluctuation component.

[0053] A phased risk assessment map is generated by evaluating the dynamic blood pressure change trajectory based on abnormal fluctuation intervals. The start and end times of the abnormal fluctuation intervals are precisely located on the three-stage time axis defined by the phase classification of the dynamic blood pressure change trajectory. The significance scores of the abnormal fluctuation intervals within each stage are calculated according to the stage classification. The risk level of each stage is divided into three levels: low risk, medium risk, and high risk based on the total score. A total score below 30 mmHg·s is low risk, 30 to 60 is medium risk, and above 60 is high risk. The same classification standard is used for all three stages. The trajectory deviation value sequence of the dynamic blood pressure change trajectory is used to extract the extreme deviation values ​​within the start and end times of the abnormal fluctuation intervals. The extreme deviation values ​​and significance scores together constitute a description of the risk intensity of that interval. The more negative the extreme value and the higher the significance score, the more severe the hypotension event. After ranking the significance scores of the abnormal fluctuation intervals within each stage, the start and end times of the top three high-scoring intervals are marked as core risk times. The core risk times are included in the phased risk assessment map in the form of precise timestamps. When patients with orthostatic hypotension change from a supine to an upright position, the significance scores during the transition period are concentrated and high. The risk level during this transition period is typically marked as high risk on the stage risk assessment chart. Conversely, if blood pressure effectively recovers during the stable period, the stage risk assessment chart is marked as low risk. The comparison of risk levels between the two stages intuitively reflects the adjustment and recovery effect. The stage risk assessment chart primarily stores the risk level of each stage and the core risk time. The risk level of each stage is determined by summarizing the significance scores of the abnormal fluctuation intervals, while the core risk time is determined by the start and end times of the abnormal fluctuation intervals and the ranking of significance scores.

[0054] Step S150: Based on the stage risk assessment map and the blood pressure baseline curve, perform adaptive fusion of body position to determine the differentiated early warning threshold, and output the blood pressure change assessment result based on the determination of exceeding the limit according to the differentiated early warning threshold.

[0055] In some embodiments, determining a differentiated early warning threshold by adaptive fusion of body position based on the stage risk assessment map and the blood pressure baseline curve includes: performing inter-stage risk gradient analysis on the stage risk assessment map to extract body position change direction features; performing a direction sensitivity assessment on the blood pressure baseline curve based on the body position change direction features to generate a direction adaptive fusion weight; performing a weighted fusion of the stage risk assessment map and the blood pressure baseline curve based on the direction adaptive fusion weight to generate a fusion assessment index; and determining a differentiated early warning threshold based on the fusion assessment index.

[0056] Inter-stage risk gradient analysis is performed on the stage risk assessment map to extract the directional features of body position transitions. The inter-stage risk gradient is defined as the difference in risk level values ​​between adjacent stages in the stage risk assessment map, calculated as G_ij = L_j − L_i, where G_ij is the risk gradient from stage i to stage j, and L_i and L_j are the risk level values ​​for stages i and j, respectively (1 for low risk, 2 for medium risk, and 3 for high risk). A G_ij greater than 0 indicates a positive gradient, and a G_ij less than 0 indicates a negative gradient. The difference between the high-risk level during the transition period and the low-risk level during the preparation period in the stage risk assessment map forms a positive gradient component. This component is typically higher during the transition from supine to standing position than during the transition from standing to supine position. The magnitude of the positive gradient component is the core basis for determining the directional features of body position transitions. Inter-stage risk gradient analysis synchronously calculates the gradient from the transition period to the stable period in the stage risk assessment map. A negative gradient indicates a decrease in risk after entering the stable period. The absolute value of the negative gradient reflects the speed of blood pressure regulation recovery; patients with slow recovery have smaller differences between the stable and transition period levels in the stage risk assessment map. The core risk moment in the stage risk assessment map participates in the construction of the temporal component of the postural transition direction feature. The delay between the core risk moment and the trigger moment of the postural transition event reflects the directional lag characteristic of the blood pressure response; the delay from supine to standing position is usually longer than that from standing to supine position. The postural transition direction feature is jointly described by three components: the gradient from the preparation period to the transition period, the gradient from the transition period to the stable period, and the temporal delay of the core risk moment. The postural transition direction feature includes a directional gradient value and a transition direction identifier. The directional gradient value is calculated from the risk level difference of each stage in the stage risk assessment map, and the transition direction identifier is inherited from the transition direction inheritance field of the stage division identifier.

[0057] For example, the step of evaluating the directional sensitivity of the blood pressure baseline curve based on the body position change direction features to generate directional adaptive fusion weights includes: classifying the body position change direction features to determine the body position change type; obtaining the corresponding clinical reference sensitivity interval based on the body position change type mapping; using the clinical reference sensitivity interval to perform individualized sensitivity correction on the blood pressure baseline curve to generate corrected sensitivity parameters; and determining the directional adaptive fusion weights based on the corrected sensitivity parameters.

[0058] The type of postural transition is determined by classifying the directional features of the postural transition. The directional gradient value of the postural transition directional features includes a positive gradient component from the preparation phase to the transition phase and a negative gradient component from the transition phase to the stabilization phase. Classification is performed by pattern matching of the signs of the two gradient components. A combination with a high positive component and a large absolute value of the negative component corresponds to a typical high-risk distribution pattern for transitions from supine to standing positions. The transition direction identifier of the postural transition directional features serves as the initial classification basis. Classification is further improved by cross-validating the transition direction identifier with the risk pattern of the directional gradient value. If the two are inconsistent, the pattern determination result of the directional gradient value takes precedence. When the positive component of the directional gradient value in the postural transition directional features exceeds one risk level and the absolute value of the negative component is less than 0.5 risk levels, the postural transition type is determined to be a high-risk type from supine to standing positions. This type corresponds to severe orthostatic hypotension with slow recovery after a sudden drop in blood pressure. The classification results of postural transition types are expressed by type labels, which are categorized into four types: standard supine to standing, high-risk supine to standing, standard standing to supine, and delayed recovery standing to supine. The distinction between each type is based on a combined determination of the directional gradient value of the postural transition direction characteristics and the transition direction label. In elderly patients with impaired autonomic function, the positive gradient component of the postural transition direction characteristics is usually larger, and the postural transition type tends to be classified as high-risk, corresponding to the clinical pattern of a significantly increased incidence of hypotension when transitioning from supine to standing in this type of patient.

[0059] The clinical reference sensitivity intervals are obtained based on the mapping of body position transformation types. The type identifier of the body position transformation type serves as the query key to perform a mapping query in the clinical reference sensitivity database. Each type identifier in the database corresponds to a preset range of systolic blood pressure sensitivity values, and each of the four type identifiers has its own dedicated clinical reference sensitivity interval entry. The values ​​of the clinical reference sensitivity intervals are determined based on large-sample clinical studies of orthostatic hypotension. The clinical reference sensitivity interval corresponding to the high-risk type from supine to standing position is 0.65 to 0.85. These patients experience a large drop in blood pressure upon standing, and the baseline curve's contribution weight to the early warning calculation needs to be correspondingly higher. The upper limit of the interval, 0.85, is a numerical representation of this clinical pattern. When the body position transformation type identifier is the standard type from supine to standing position, the clinical reference sensitivity interval is 0.45 to 0.65, corresponding to a moderate risk of sudden blood pressure drop and a moderate impact of baseline levels on the threshold setting. The upper and lower bounds of the clinical reference sensitivity interval together constitute the value constraint range for individualized sensitivity correction. The correction result must not exceed the upper and lower bounds determined by the clinical reference sensitivity interval to ensure that the individualized result is within the clinically safe range. When database mapping fails, the four type identifiers for position conversion are filled with the clinical reference sensitivity interval of the standard supine to standing position type as the default value. The clinical reference sensitivity interval includes an upper and lower sensitivity bound, both of which are determined by the type identifier of the position conversion type through a database mapping query.

[0060] The blood pressure baseline curve is individually sensitively corrected using a clinical reference sensitivity interval to generate a corrected sensitivity parameter. When the mean of the baseline blood pressure value sequence during the transition period is below 110 mmHg, the individualized sensitivity correction pushes the corrected sensitivity parameter towards the upper bound of the clinical reference sensitivity interval. Patients with low baselines have insufficient blood pressure reserve; even a slight drop after standing can trigger symptoms, requiring the blood pressure baseline curve to contribute more to the warning threshold. Pushing it to the upper bound reflects this clinical logic. The width of the clinical reference sensitivity interval is proportional to the standard deviation of the blood pressure baseline curve during the transition period, adjusting the correction step size. When the blood pressure baseline curve fluctuates significantly during the transition period, the correction step size is smaller to maintain the robustness of the correction results. When the peak-to-trough difference of the baseline blood pressure value sequence during the transition period exceeds 20 mmHg, the corrected sensitivity value of the corrected sensitivity parameter is determined by shifting it upwards by 0.1 from the median of the clinical reference sensitivity interval, reflecting the need to amplify the warning fusion weight for high-fluctuation baselines. The upper bound constraint of the clinical reference sensitivity interval ensures that the corrected sensitivity value of the sensitivity parameter does not exceed the safety upper limit, while the lower bound constraint ensures that the corrected sensitivity value is not lower than the minimum effective threshold, preventing insufficient contribution of the blood pressure baseline curve to the fusion results. The corrected sensitivity parameter includes the corrected sensitivity value and the corrected confidence level. The corrected sensitivity value is calculated by mapping the statistical characteristics of the baseline blood pressure value sequence of the blood pressure baseline curve within the clinical reference sensitivity interval. The corrected confidence level is jointly evaluated by the interval width of the clinical reference sensitivity interval and the number of effective sampling points of the blood pressure baseline curve.

[0061] The directional adaptive fusion weights are determined based on the corrected sensitivity parameter. The corrected sensitivity value of the parameter directly serves as the core numerical source for the directional adaptive fusion weights. A corrected sensitivity value between 0.65 and 0.85 corresponds to a higher value range for the directional adaptive fusion weights, reflecting the high contribution of the blood pressure baseline curve to the fusion results at that sensitivity level. The corrected confidence level of the parameter is used to apply confidence weighting to the directional adaptive fusion weights. A high confidence level results in the directional adaptive fusion weights directly using the corrected sensitivity value; a low confidence level—such as insufficient effective patient sampling points leading to instability in the baseline curve statistics—results in the directional adaptive fusion weights shrinking towards the median of the clinical reference sensitivity interval, reducing the excessive influence of low-confidence correction results on the fusion calculation. The fusion weight value is calculated by multiplying the corrected sensitivity value by the corrected confidence level normalization coefficient. Normalization ensures that the weight value falls within the constraint range of 0.4 to 0.9. The directional adaptive fusion weights for high-risk transitions are typically higher than those for standard transitions, giving the blood pressure baseline curve for high-risk transitions a greater weight in the fusion. The directional adaptive fusion weights consist of fusion weight values ​​and a sequence of valid weight timestamps. The fusion weight values ​​are jointly determined by the corrected sensitivity value and corrected confidence level of the corrected sensitivity parameter, while the valid weight timestamp sequence indicates the effective time range of the weight values.

[0062] Based on the directional adaptive fusion weight, the stage risk assessment map and the blood pressure baseline curve are weighted and fused to generate a fusion assessment index. The fusion weight value of the directional adaptive fusion weight ranges from 0.4 to 0.9. A higher fusion weight value indicates a greater contribution of the blood pressure baseline curve to the fusion result. The effective timestamp sequence of the directional adaptive fusion weight marks the effective time range of the weighted fusion. The weighted fusion is performed time-by-time within this time range. The calculation formula is FI(t) = BP_base(t) − w × R_norm(t) × D_ref, where FI(t) is the fusion evaluation index value at time t (in mmHg), BP_base(t) is the baseline blood pressure value of the blood pressure baseline curve at time t (in mmHg), w is the fusion weight value of the directional adaptive fusion weight (dimensionless), R_norm(t) is the normalized value of the risk level corresponding to the stage risk assessment map (0.2 for low risk, 0.5 for medium risk, and 0.8 for high risk), and D_ref is the absolute value of the difference between the mean of the blood pressure baseline curve during the preparation period and 90 mmHg (in mmHg). A higher risk normalization value results in a greater downward adjustment of the fusion evaluation index. For patients with a fusion weight of 0.5, the mean blood pressure baseline during the preparation phase was 117 mmHg, corresponding to a D_ref of 27 mmHg. During the low-risk preparation phase, the fusion assessment index was approximately 114, decreasing to approximately 106 during the high-risk transition phase. This difference between the two phases directly forms the basis for calculating the differential warning threshold. The fusion assessment index is organized and stored as a sequence of fusion index values ​​and an index timestamp sequence, which are used for quantile calculation and time axis alignment of the differential warning threshold, respectively.

[0063] Differential warning thresholds are determined based on the fusion assessment index. The distribution characteristics of the fusion index value sequence within each stage determine the stage-specific variation of the differential warning threshold. The greater the difference between the minimum value of the fusion index sequence during the transition period and the minimum value during the preparation period, the greater the tightening of the differential warning threshold during the transition period relative to the preparation period. The differential warning threshold is set based on the 10th percentile of the fusion index value sequence in each stage. The 10th percentile corresponds to the lower end of the fusion index distribution in each stage, and the differential warning threshold based on this can cover more than 90% of abnormally low events. If the fusion index value sequence during the stable period of the fusion assessment index remains below 80, it indicates that the patient's blood pressure has not effectively recovered to the baseline level under the new position. The stable period component of the differential warning threshold will be tightened to the same level as the transition period, maintaining a consistently strict warning standard for such patients with delayed adjustment. The index timestamp sequence of the fusion assessment index is used to map each stage component of the differential warning threshold to the time axis. The effective period of the stage warning threshold is determined by the corresponding time range of each stage of the fusion assessment index. The directional difference coefficient of the differentiated warning threshold is obtained by statistically analyzing the difference between the fusion assessment index values ​​in the patient's historical positional change records for the two directions: supine to standing and standing to supine. When historical records are insufficient, the directional difference coefficient uses the population default value of 1.15. The directional difference coefficient quantifies the degree of asymmetry in warning sensitivity between the two change directions. The differentiated warning threshold includes the warning threshold for each stage and the directional difference coefficient. The warning threshold for each stage is determined by the 10th percentile of the fusion assessment index value sequence for each stage, and the directional difference coefficient is calculated by statistically analyzing the difference between the fusion assessment index values ​​in the two change directions.

[0064] The blood pressure change assessment results are output based on the differential warning threshold. The differential warning threshold is used as a comparison benchmark, and the stroke-weighted systolic blood pressure of the real-time blood pressure sample is judged point by point. The warning threshold at each stage of the differential warning threshold provides a dynamic comparison benchmark for point-by-point judgment within each stage. When the stroke-weighted systolic blood pressure is continuously lower than the corresponding stage warning threshold for more than 3 sampling points, an over-limit marker is triggered. The directional difference coefficient of the differential warning threshold is used for directional correction of the over-limit judgment. The directional correction threshold is obtained by multiplying the warning threshold at each stage of the differential warning threshold during the transition from supine to standing position by the directional difference coefficient, reflecting a stricter warning standard for orthostatic hypotension during the transition from supine to standing position. In high-risk patients with orthostatic hypotension, the directional correction threshold during the transition period is usually lower than 90 mmHg systolic blood pressure. When the real-time blood pressure sample value continuously falls below this threshold, the blood pressure change assessment result is marked as an orthostatic hypotension event; when the blood pressure fails to recover to above the stable period warning threshold during the stable period, the blood pressure change assessment result is marked as a delayed blood pressure recovery event; when the oscillation amplitude continuously exceeds 5 mmHg above or below the threshold during the transition period, the blood pressure change assessment result is marked as an increased blood pressure variability event. The blood pressure change assessment results are output to the clinic in four aspects: event type, time of exceeding the limit, magnitude of exceeding the limit, and stage classification. The magnitude of exceeding the limit is defined as the absolute value of the difference between the real-time blood pressure sample value and the differentiated warning threshold, and is summarized and output in the form of a structured report.

[0065] To implement the above-described method embodiment, a dynamic blood pressure assessment method based on body position transition phase is provided to achieve the corresponding functional and technical effects. See also... Figure 2 , Figure 2 This paper illustrates a structural block diagram of a blood pressure dynamic assessment device 200 based on body position transition phases, according to an embodiment of this application. The device includes: Data acquisition module 201 is used to acquire patient body position sensor data and blood pressure sensor data, identify body position transformation events by performing body position transformation identification on the body position sensor data, and perform response lag compensation generation stage division and identification based on the body position transformation events. The baseline establishment module 202 is used to sample the blood pressure sensing data in stages through the stage division identifier to generate a multi-stage blood pressure sampling sequence, perform individualized baseline construction on the multi-stage blood pressure sampling sequence to generate a blood pressure baseline curve, and establish a stage comparison mapping between the blood pressure baseline curve and the real-time blood pressure sampling value to form a stage reference baseline. The deviation tracking module 203 is used to calculate the deviation of the real-time blood pressure sampling value point by point according to the stage reference benchmark to generate a blood pressure deviation sequence, detect the continuous downward trend and recovery characteristics after the decline of the blood pressure deviation sequence to generate rate constraint parameters, and perform rate constraint on the blood pressure deviation sequence according to the rate constraint parameters to form a dynamic blood pressure change trajectory. The risk assessment module 204 is used to separate the physiological rhythm fluctuations of the blood pressure dynamic change trajectory to determine the abnormal fluctuation range, and to perform risk assessment on the blood pressure dynamic change trajectory based on the abnormal fluctuation range to form a stage risk assessment map. The early warning output module 205 is used to determine a differentiated early warning threshold by adaptively fusing the stage risk assessment map and the blood pressure reference curve according to body position, and to output the blood pressure change assessment result based on the differential early warning threshold.

[0066] The aforementioned blood pressure dynamic assessment device 200 based on the body position transition phase can implement one of the blood pressure dynamic assessment methods based on the body position transition phase described in the above method embodiments. The options in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0067] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

Claims

1. A method for dynamic blood pressure assessment based on body position transition phases, characterized in that, include: Collect patient posture sensor data and blood pressure sensor data, identify posture transition events by performing posture transition recognition on the posture sensor data, and divide and identify the response lag compensation generation stage based on the posture transition events. The blood pressure sensor data is sampled in stages using the stage division identifier to generate a multi-stage blood pressure sampling sequence. An individualized baseline is constructed from the multi-stage blood pressure sampling sequence to generate a blood pressure reference curve. The blood pressure reference curve is compared and mapped with the real-time blood pressure sampling value to form a stage reference benchmark. Based on the stage reference benchmark, the deviation of the real-time blood pressure sampling value is calculated point by point to generate a blood pressure deviation sequence. The continuous downward trend and recovery characteristics after the decline are detected in the blood pressure deviation sequence to generate rate constraint parameters. Based on the rate constraint parameters, the blood pressure deviation sequence is rate-constrained to form a dynamic blood pressure change trajectory. The abnormal fluctuation range is determined by separating the physiological rhythm fluctuations of the blood pressure dynamic change trajectory, and a staged risk assessment map is formed based on the abnormal fluctuation range of the blood pressure dynamic change trajectory. Based on the stage risk assessment map and the blood pressure baseline curve, a differentiated early warning threshold is determined by adaptive fusion of body position. Based on the differentiated early warning threshold, the blood pressure change assessment result is output when the threshold is exceeded.

2. The method according to claim 1, characterized in that, The generation stage segmentation identifier based on the response lag compensation generated by the body position change event includes: Based on the aforementioned body position change event, a response delay estimation is performed to generate a pressure receptor reflection delay window; Individualized calibration of the pressure sensor reflection delay window generates calibration delay parameters; The calibration delay parameter is used to perform time compensation on the body position transition event to generate a hysteresis compensation stage boundary. The stage division identifiers are determined based on the boundaries of the delayed compensation stage, which define the preparation period, transition period, and stabilization period.

3. The method according to claim 1, characterized in that, The step of constructing an individualized baseline for the multi-stage blood pressure sampling sequence to generate a blood pressure baseline curve includes: The resting-state stability of the multi-stage blood pressure sampling sequence is screened to generate a stability qualification mark; Based on the stability qualification marker, outlier sampling points are identified by performing outlier detection on the multi-stage blood pressure sampling sequence. The outlier sampling points are excluded and corrected to generate a corrected sampling sequence. The corrected sampling sequence is used to perform multi-stage segmented fitting to generate a blood pressure baseline curve.

4. The method according to claim 1, characterized in that, The generation rate constraint parameters for detecting the continuous downward trend and recovery characteristics after the decrease in the blood pressure deviation sequence include: Time-segment analysis was performed on the blood pressure deviation sequence to identify continuously decreasing deviation segments; For the continuous downward deviation segment, perform valley location and recovery inflection point identification to generate a downward-recovery time window; Based on the descent-recovery time window, the rate features of the continuous descent deviation segment are extracted to generate a segmented rate threshold. Rate constraint parameters are determined based on the segmented rate threshold.

5. The method according to claim 1, characterized in that, The step of separating abnormal fluctuation ranges from the physiological rhythm fluctuations of the blood pressure dynamic change trajectory includes: Local mean deviation analysis is performed on the dynamic blood pressure change trajectory to generate a fluctuation residual sequence; Individualized respiratory cycle estimation is performed on the fluctuating residual sequence to generate an adaptive rhythm template; Based on the adaptive rhythm template, template matching is performed on the fluctuation residual sequence to separate and extract non-rhythmic fluctuation components; The abnormal fluctuation range is determined based on the amplitude and significance score of the non-rhythmic fluctuation component.

6. The method according to claim 1, characterized in that, The step of determining differentiated early warning thresholds by adaptively fusing body position based on the stage risk assessment map and the blood pressure baseline curve includes: The risk gradient analysis between stages of the stage risk assessment map is performed to extract the positional transformation direction features; Based on the body position change direction characteristics, the blood pressure baseline curve is subjected to direction sensitivity evaluation to generate direction adaptive fusion weights; Based on the aforementioned directional adaptive fusion weights, a weighted fusion of the stage risk assessment map and the blood pressure baseline curve is performed to generate a fusion assessment index; The differentiated early warning threshold is determined based on the fusion evaluation index.

7. The method according to claim 1, characterized in that, The step of establishing a phased reference benchmark by comparing and mapping the blood pressure baseline curve with real-time blood pressure sampling values ​​includes: Blood pressure variability is assessed on the blood pressure baseline curve and the real-time blood pressure sampling values ​​to extract the variability level corresponding to each stage; Configure phased weighting coefficients based on the degree of variability corresponding to each phase; Based on the phased weighting coefficients, a weighted ratio is performed on the blood pressure baseline curve and the real-time blood pressure sampling value to generate a weighted deviation component; The stage weight mapping table is constructed using the weighted deviation components to generate a stage reference benchmark.

8. The method according to claim 5, characterized in that, The step of generating an adaptive rhythm template by estimating the individualized respiratory cycle for the fluctuating residual sequence includes: Periodic analysis is performed on the fluctuation residual sequence to generate a residual periodic distribution map; Based on the residual period distribution map, the main respiratory wave period and the main Mayer wave period are extracted to determine the reference period of the dual-band physiological rhythm. Based on the dual-band physiological rhythm reference period, adaptive periodic tracking is performed on the fluctuation residual sequence to generate dynamic rhythm waveforms; An adaptive rhythm template is constructed based on the dynamic rhythm waveform.

9. The method according to claim 6, characterized in that, The step of evaluating the direction sensitivity of the blood pressure baseline curve based on the body position change direction characteristics to generate direction-adaptive fusion weights includes: The body position transition direction features are categorized to determine the body position transition type; The corresponding clinical reference sensitivity interval is obtained based on the body position conversion type mapping; The blood pressure baseline curve is individually sensitively corrected using the clinical reference sensitivity range to generate corrected sensitivity parameters; The directional adaptive fusion weights are determined based on the modified sensitivity parameters.

10. A device for dynamic blood pressure assessment based on body position transition phase, characterized in that, include: The data acquisition module is used to collect patient posture sensor data and blood pressure sensor data, identify posture transition events by performing posture transition identification on the posture sensor data, and perform response lag compensation generation stage segmentation based on the posture transition events. The baseline establishment module is used to sample the blood pressure sensing data in stages according to the stage division identifier to generate a multi-stage blood pressure sampling sequence, perform individualized baseline construction on the multi-stage blood pressure sampling sequence to generate a blood pressure baseline curve, and establish a stage comparison mapping between the blood pressure baseline curve and the real-time blood pressure sampling value to form a stage reference baseline. The deviation tracking module is used to calculate the deviation of the real-time blood pressure sampling value point by point according to the stage reference benchmark to generate a blood pressure deviation sequence, detect the continuous downward trend and recovery characteristics after the decline of the blood pressure deviation sequence to generate rate constraint parameters, and perform rate constraint on the blood pressure deviation sequence according to the rate constraint parameters to form a dynamic blood pressure change trajectory. The risk assessment module is used to separate the physiological rhythm fluctuations of the blood pressure dynamic change trajectory to determine the abnormal fluctuation range, and to perform risk assessment on the blood pressure dynamic change trajectory based on the abnormal fluctuation range to form a stage risk assessment map. The early warning output module is used to determine a differentiated early warning threshold by adaptively fusing the stage risk assessment map and the blood pressure baseline curve according to body position, and to output the blood pressure change assessment result based on the differentiated early warning threshold to determine if the threshold is exceeded.

Images