An icu rehabilitation decision support system based on causal inference
By quantifying the synchronicity between physiological subsystems in the ICU and using causal inference methods to assess patients' tolerance to rehabilitation treatment, the problem of insufficient real-time performance and reliability in existing technologies is solved, and accurate decision support with low latency and low resource consumption is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST AFFILIATED HOSPITAL ZHEJIANG UNIV COLLEGE OF MEDICINE
- Filing Date
- 2026-02-05
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies for rehabilitation treatment decision support in the ICU suffer from poor real-time performance, high computational complexity, and insufficient reliability. They are unable to accurately assess patients' tolerance to rehabilitation treatment, leading to frequent false alarms and missed alarms, and thus failing to meet the need for real-time and efficient bedside decision support.
Using a causal inference-based approach, the system acquires multiple physiological time-series signals from patients, quantifies the synchronicity between physiological subsystems, calculates the perturbation tolerance index, and compares it with the preset rehabilitation intervention perturbation equivalent value to provide immediate decision support.
It enables real-time decision support with low latency and low resource consumption, improving the accuracy and reliability of decision-making, providing intuitive and interpretable clinical guidance, and reducing false alarms and missed alarms.
Smart Images
Figure CN122158111A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical information processing technology, and more specifically, to a system and method for rehabilitation treatment decision support in an intensive care unit (ICU) setting. Background Technology
[0002] Once the vital signs of patients in the Intensive Care Unit (ICU) have stabilized, early implementation of rehabilitation therapy is crucial for improving prognosis and reducing the incidence of complications. However, as an externally applied medical procedure, rehabilitation intervention may pose certain risks to patients whose physiological state is not yet fully stable. Therefore, accurately assessing a patient's tolerance to specific rehabilitation treatments at specific time points has become a critical technical issue that urgently needs to be addressed in clinical practice.
[0003] Currently, most common existing technologies rely on static thresholds of single physiological indicators such as heart rate and blood pressure for alarm and judgment. Although this method is simple and easy to implement, it is difficult to reflect the complex dynamic coupling relationship between multiple physiological systems, which can easily lead to false alarms and false alarms, affecting the accuracy and reliability of decision-making. Some other solutions attempt to improve the scientific nature of judgment by constructing complex probabilistic graphical models for causal inference, but they often have limitations such as large computational delay, high resource consumption, and poor adaptability to new patients, making it difficult to meet the needs of real-time and efficient decision support at the ICU bedside. The existing problems mainly include: (1) the static method of single physiological indicators cannot fully analyze the multi-system synergy and compensation mechanism; (2) the complex causal model is limited by real-time performance and generalization ability, making it difficult for existing technologies to achieve reliable real-time decision support in actual ICU scenarios. In this context, how to achieve low-latency, low-resource consumption and rapid adaptation to new patients while ensuring the accuracy of causal inference has become a technical bottleneck that needs to be overcome in this field. This application aims to provide a decision support solution that is suitable for the ICU bedside environment and can assess the tolerance of rehabilitation treatment in real time and accurately, providing a scientific tool for critical care rehabilitation practice. Summary of the Invention
[0004] The purpose of this application is to provide an ICU rehabilitation decision support system and method based on causal inference, which aims to solve the technical problems of poor real-time performance, high computational complexity and insufficient reliability in the existing rehabilitation decision support technology.
[0005] To achieve the above objectives, in a first aspect, this application provides an ICU rehabilitation decision support system based on causal inference, comprising: a data acquisition and preprocessing module configured to acquire at least two real-time physiological time-series signals from a patient; a physiological rhythm synchronicity analysis module configured to determine a perturbation tolerance index characterizing the dynamic stability of the patient's physiological system based on the transspectral coherence between the at least two physiological time-series signals; a rehabilitation intervention database module configured to store multiple preset perturbation equivalent values corresponding to various rehabilitation intervention measures, and to provide a perturbation equivalent value corresponding to the rehabilitation intervention measure to be executed indicated by a rehabilitation intervention selection instruction; and a risk assessment and decision support module configured to generate decision support information suggesting the execution of the rehabilitation intervention measure to be executed when the perturbation tolerance index is greater than the perturbation equivalent value. This application assesses the stability of the overall system by directly quantifying the synchronicity between physiological subsystems and transforms the complex risk assessment problem into a direct comparison of two scalar values, thereby achieving real-time decision support with low latency and low computational power.
[0006] Optionally, the at least two physiological time series signals include a heart rate variability time series and a respiratory rate variability time series.
[0007] Optionally, the data acquisition and preprocessing module is specifically configured to: synchronously acquire electrocardiogram (ECG) signals and respiratory signals; generate the heart rate variability time series based on the R-wave interval of the ECG signals; and generate the respiratory rate variability time series based on the respiratory cycle of the respiratory signals.
[0008] Optionally, the physiological rhythm synchronization analysis module is specifically configured to: perform spectral analysis on the at least two physiological time series signals to obtain their respective power spectral density estimates and mutual power spectral density estimates; calculate an amplitude squared coherence function within a preset frequency band based on the power spectral density estimates and the mutual power spectral density estimates; and determine the disturbance tolerance index based on the integral or mean of the amplitude squared coherence function within the preset frequency band.
[0009] Optionally, the preset frequency band is 0.15 Hz to 0.4 Hz.
[0010] Optionally, the risk assessment and decision support module is further configured to generate decision support information for temporarily suspending the rehabilitation intervention measures to be implemented when the disturbance tolerance index is less than or equal to the disturbance equivalent value.
[0011] Secondly, this application provides an ICU rehabilitation decision support method based on causal inference, comprising: acquiring at least two real-time physiological time-series signals from the patient; determining a disturbance tolerance index characterizing the dynamic stability of the patient's physiological system based on the transspectral coherence between the at least two physiological time-series signals; acquiring a preset disturbance equivalent value corresponding to a rehabilitation intervention to be performed; and generating decision support information suggesting the execution of the rehabilitation intervention to be performed when the disturbance tolerance index is greater than the disturbance equivalent value. This method has a clear flow, direct calculation, and can be efficiently deployed in bedside monitoring equipment, providing clinicians with immediate and interpretable decision-making basis.
[0012] Optionally, the at least two physiological time series signals include a heart rate variability time series and a respiratory rate variability time series.
[0013] Optionally, the step of determining a disturbance tolerance index characterizing the dynamic stability of a patient's physiological system based on the transspectral coherence between the at least two physiological time-series signals includes: performing spectral analysis on the at least two physiological time-series signals to obtain their respective power spectral density estimates and mutual power spectral density estimates; calculating an amplitude squared coherence function within a preset frequency band based on the power spectral density estimates and the mutual power spectral density estimates; and determining the disturbance tolerance index based on the integral or mean of the amplitude squared coherence function within the preset frequency band.
[0014] Optionally, the preset frequency band is 0.15 Hz to 0.4 Hz.
[0015] The beneficial effects of this application are as follows: by introducing physiological system perturbation tolerance and rehabilitation intervention perturbation equivalent, the complex problem of physiological state assessment is transformed into a deterministic engineering problem with low computational complexity. No model training is required, there is no cold start problem, and stable and reliable decision support can be provided from the start of data acquisition. Decision latency can be controlled at the millisecond level, improving the real-time performance of decisions. At the same time, its decision logic (comparing the magnitudes of two quantitative indicators) is intuitive and highly interpretable, helping to build trust among clinicians. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1This is a structural block diagram of an ICU rehabilitation decision support system based on causal inference provided in one embodiment of this application.
[0018] Figure 2 This is a flowchart of an ICU rehabilitation decision support method based on causal inference provided in one embodiment of this application.
[0019] Figure 3 This is a schematic diagram illustrating the calculation process of the Physiological System Disturbance Tolerance (PSPT) index in one embodiment of this application.
[0020] Figure 4 This is a schematic diagram of the calibration process for the Rehabilitation Intervention Perturbation Equivalent (RIPE) value in one embodiment of this application.
[0021] Figure 5 This is a conceptual diagram used to illustrate the core technical principles of this application, showing the characteristics of HRV and RRV signals in the time and frequency domains and their interrelationships; the upper left and upper right figures show the waveforms of the HRV and RRV signals in the time domain, respectively; the lower left figure shows the power spectral density distribution of the two signals in the frequency domain; the lower right figure shows the amplitude squared coherence function between the two signals, and highlights the target frequency band used to calculate the PSPT index. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0023] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0024] Example 1
[0025] Reference Figure 1This embodiment provides an ICU rehabilitation decision support system 100 based on causal inference. In a specific implementation, this system assesses the overall stability of the patient as a complex dynamic system by quantifying the coupling strength between physiological subsystems and comparing it with a preset rehabilitation intervention stress level. This provides real-time, prospective risk assessment for rehabilitation decisions. This method solves the technical problems of inaccurate assessments due to reliance on static threshold alarms and high computational latency and inability to be applied immediately due to reliance on complex models in existing technologies, thus improving the safety and scientific rigor of ICU rehabilitation decisions.
[0026] The system 100 includes a data acquisition and preprocessing module 110, a physiological rhythm synchronization analysis module 120, a rehabilitation intervention database module 130, and a risk assessment and decision support module 140.
[0027] The data acquisition and preprocessing module 110 is configured to acquire at least two real-time physiological time-series signals from the patient. In a specific implementation scenario, this module connects to various monitoring devices in the ICU ward, such as electrocardiogram monitors, ventilators, and pulse oximeters, via standard medical device communication protocols, such as HL7 (HealthLevel Seven) or software development kits (SDKs) provided by specific device manufacturers. One of the core functions of this module is to ensure that signals acquired from different physical sources are precisely aligned in the time dimension, which can be synchronized through a unified system clock or Network Time Protocol (NTP). The acquired data is typically raw, high-sampling-rate waveform data. For subsequent analysis, the module also needs to perform a series of preprocessing operations.
[0028] For example, suppose we need to acquire time series data of heart rate variability and respiratory rate variability. The data acquisition and preprocessing module 110 first acquires electrocardiogram (ECG) signals at a sampling rate of 250 Hz and respiratory airflow signals at a sampling rate of 50 Hz. For the acquired ECG signals, the signal processing unit within the module applies a bandpass filter, for example, with a passband range of 0.5 Hz to 40 Hz, to filter out baseline drift and electromyographic noise. After filtering, a QRS complex detection algorithm, such as Pan-Tompkins, is used to identify the R-wave peaks in each heartbeat cycle and record their timestamps. By calculating the time intervals between consecutive R-wave peaks, a jump-by-jump RR interval (RRI) sequence is obtained. Similarly, for the respiratory airflow signals, the module identifies the peaks of inspiration and expiration to determine the duration of each respiratory cycle, forming a jump-by-jump respiratory cycle length (BBI) sequence. Since both the RRI and BBI sequences are sampled at non-uniform time intervals, they must be converted into time series with uniform time intervals for frequency domain analysis. This module uses a cubic spline interpolation algorithm to resample the RRI and BBI sequences, uniformly converting them into time series with a sampling rate of 4 Hz, thereby generating heart rate variability (HRV) time series. and respiratory rate variability (RRV) time series Finally, the module divides the two processed time-series signals into a fixed data window length, such as 256 sampling points (corresponding to a duration of 64 seconds), and packages these data windows as output, transmitting them to the physiological rhythm synchronicity analysis module 120. This 64-second data window length is the result of a technical balance between ensuring that each data window contains at least six complete respiratory cycles to obtain stable spectral estimation (this is the frequency resolution requirement) and maintaining a rapid response capability to changes in the patient's physiological state (this is the time resolution requirement).
[0029] The physiological rhythm synchronization analysis module 120 is configured to determine a disturbance tolerance index characterizing the dynamic stability of a patient's physiological system based on the transspectral coherence between the at least two physiological time series signals.
[0030] To more intuitively demonstrate the signal characteristics analyzed in this application and their core analytical principles, please refer to [link / reference needed]. Figure 5 . Figure 5 The diagram schematically illustrates the time-domain morphology of HRV and RRV signals, their respective power spectral distributions in the frequency domain, and the transspectral coherence function between them. For example... Figure 5 As shown in the lower right sub-figure, the degree of synchronization between two physiological rhythms can be quantified by calculating the coherence intensity within a preset target frequency band (such as the highlighted area in the figure).
[0031] This application defines this indicator as Physiological System Perturbation Tolerance (PSPT). The physical meaning of this indicator lies in quantifying the degree of coordination and synchronization among key subsystems within a patient's physiological system (such as the cardiovascular and respiratory systems). A highly synchronized and coordinated system has a stable internal structure and a strong ability to resist external disturbances, thus exhibiting high perturbation tolerance. Conversely, when a system is on the verge of instability, the coupling between subsystems weakens first, manifesting as a decrease in synchronicity, and consequently, a decrease in perturbation tolerance. Therefore, the PSPT value becomes an indicator for measuring the intrinsic stability of a system.
[0032] Figure 3 The complete calculation process of the PSPT index is shown in detail. The process starts with acquiring two physiological time series signals as input, and goes through windowing, FFT transformation, power spectrum and cross-power spectrum estimation, coherence calculation, and finally integration or averaging within a specific frequency band, and then scaling to obtain the final PSPT value.
[0033] For example, the circadian rhythm synchronicity analysis module 120 receives a data window from module 110 containing a 256-point HRV sequence and a 256-point RRV sequence. First, to reduce spectral leakage during subsequent Fourier transforms, the module applies a Hanning window function to each of the two sequences. Then, the module performs a Fast Fourier Transform (FFT) on the windowed sequences to obtain their complex representations in the frequency domain. Based on this, the module can calculate their respective power spectral density (PSD) estimates. and And their cross-power spectral density (CPSD) estimation These calculations can be performed using the Welch average periodogram method to obtain smoother and more reliable spectral estimates. The amplitude squared coherence function is defined as follows: The function, with a value between 0 and 1, quantifies the degree of linear correlation between two signals at a specific frequency f. Since the modulation of heart rate by respiratory activity (i.e., respiratory sinus arrhythmia) is primarily concentrated in a specific high-frequency band, the module focuses on a preset frequency band, such as 0.15 Hz to 0.4 Hz. This band was chosen because it precisely corresponds to the physiologically recognized high-frequency power spectrum range reflecting respiratory sinus arrhythmia in healthy adults, ensuring the physiological significance of the analysis. Finally, the module obtains a single scalar by integrating or calculating the arithmetic mean of the coherence function values within this band. For example, this scalar... To facilitate understanding and use, this module normalizes this scalar, for example, by multiplying it by 100, to obtain the final PSPT value. If the calculated... If the value is 0.82, then the output PSPT value is 82.0 (dimensionless).
[0034] The rehabilitation intervention database module 130 is configured to store multiple preset perturbation equivalent values corresponding to various rehabilitation intervention measures. This application defines these preset values as Rehabilitation Intervention Perturbation Equivalent (RIPE). The RIPE value is a dimensionless calibration value that represents the magnitude of a typical "stress" or "perturbation" exerted on the patient's physiological system by a specific rehabilitation intervention.
[0035] Figure 4 The systematic calibration process for determining the RIPE value in this embodiment is illustrated in detail. As shown in the figure, the process begins with defining the assessment dimensions and inviting an expert group. The weights of each dimension are determined using the Analytic Hierarchy Process (AHP), and combined with the experts' scores for each intervention measure, an objective and repeatable RIPE value is finally calculated by weighted summation.
[0036] These values can be determined based on a systematic and repeatable calibration method.
[0037] In one embodiment, the calibration method combines the Analytic Hierarchy Process (AHP) with expert scoring. This method first defines four core dimensions for assessing the degree of disturbance caused by rehabilitation interventions: (Activity intensity, assessment of cardiovascular load) (Area of muscle groups involved) (Intervention duration) and (Patient active participation). Subsequently, an expert panel consisting of at least five senior ICU physical therapists was invited to first construct a judgment matrix using a pairwise comparison method, and then use the analytic hierarchy process (AHP) to determine the weight vectors of these four dimensions. For example, the resulting weights might be: Then, the expert panel scored each specific rehabilitation intervention (e.g., "assisting in turning over to 30 degrees") independently on a scale of 1-9 in each dimension, resulting in a score vector. Ultimately, the RIPE value of this intervention was calculated using a weighted summation: Here, K is a scaling factor used to map the final result to the [0, 200] range comparable to the PSPT value. The module can be implemented as an embedded key-value database, where the "key" is the name string of the rehabilitation intervention, and the "value" is its corresponding floating-point RIPE value. When a therapist selects a rehabilitation activity to be performed through the user interface, the module receives a query command containing the activity name and returns the corresponding RIPE value.
[0038] For example, for the intervention "assisting in rolling over to 30 degrees," the expert panel gave an average score vector of [missing information]. Using the aforementioned weight vector W and an exemplary scaling factor K=15, the RIPE value is calculated as follows: After correction and verification, the final value stored in the database may be 75.0. The entries stored in the database may be as follows: {"Intervention Name": "Passive Limb Movements in Bed", "RIPE Value": 40.0} {"Intervention Name": "Assisting in rolling over to 30 degrees", "RIPE Value": 75.0} {"Intervention Name": "Sit upright at the bedside for 5 minutes", "RIPE Value": 90.0} {"Intervention Name": "Bedside Standing Training", "RIPE Value": 120.0} When the user selects "Assist in rolling over to 30 degrees", module 130 will output a value of 75.0.
[0039] The risk assessment and decision support module 140 is configured to execute the final decision logic. It is a simple comparator that receives real-time PSPT values from the circadian rhythm synchronicity analysis module 120 and RIPE values related to user selection from the rehabilitation intervention database module 130. The core function of this module is to compare the magnitudes of these two values. Its underlying logic is: if the patient's current intrinsic stability (quantified by PSPT) exceeds the strength of the impending external perturbation (quantified by RIPE), then performing the intervention is safe. Otherwise, there is a risk.
[0040] When the disturbance tolerance index is greater than the disturbance equivalent value, the module generates decision support information suggesting the implementation of the rehabilitation intervention to be performed. For example, if module 140 receives a PSPT value of 82.0 and a RIPE value of 75.0, since 82.0 > 75.0, the module will drive the user interface to display a green indicator light or a "Suggested Implementation" text prompt.
[0041] In another optional implementation, when the perturbation tolerance index is less than or equal to the perturbation equivalent value, the module is further configured to generate decision support information to postpone the rehabilitation intervention to be performed. For example, if module 140 receives a PSPT value of 65.0, while the RIPE value of the intervention to be performed is still 75.0, since 65.0 ≤ 75.0, the module will drive the user interface to display a red warning sign or a "High risk, postponement recommended" text prompt, thereby effectively preventing rehabilitation activities that may harm the patient.
[0042] Those skilled in the art will understand that the applicability of the coupled analysis method based on heart rate variability and respiratory rate variability described in this invention may be limited in certain clinical situations. For example, for patients with fixed-frequency pacemakers, their heart rate variability is artificially eliminated; for patients using high-dose beta-blockers or other drugs affecting the autonomic nervous system, their cardiopulmonary coupling response may be significantly weakened; or for patients with severe arrhythmias (such as atrial fibrillation), the extraction of RR intervals is inherently unreliable. In these cases, the system can be configured to automatically identify such conditions (e.g., by analyzing the morphology of electrocardiogram signals or obtaining medication information from electronic medical records) and issue a warning to the user that the reliability of the current PSPT indicator may be reduced, or automatically switch to an alternative physiological signal pair for analysis, as described in Example 2 below.
[0043] Example 2 This embodiment provides another implementation of the aforementioned system. In this embodiment, the selected physiological signal pair is a blood pressure variability (BPV) time series and a heart rate variability (HRV) time series. The core physiological basis of this combination is the baroreflex, a key short-term regulatory mechanism for maintaining blood pressure stability. Decreased baroreflex sensitivity (BRS) is considered a marker of autonomic dysfunction and increased cardiovascular risk. Therefore, by quantifying the coupling strength between BPV and HRV, the integrity and effectiveness of this key negative feedback loop can be assessed, thus providing another dimension of quantitative evaluation for system stability.
[0044] In this embodiment, the structure of system 100 is the same as that of embodiment one, but the internal processing logic of data acquisition and preprocessing module 110 and physiological rhythm synchronization analysis module 120 is adjusted.
[0045] The data acquisition and preprocessing module 110 is configured to simultaneously acquire invasive or non-invasive continuous arterial blood pressure (ABP) waveform signals and electrocardiogram (ECG) signals. For the ABP signal, the module identifies the peak systolic blood pressure (SBP) in each cardiac cycle, forming a jump-by-jump SBP sequence. Simultaneously, as described in Example 1, the RR interval sequence is extracted from the ECG signal. Subsequently, these two non-equidistant sequences are similarly resampled into 4 Hz equal-interval time series using cubic spline interpolation, generating blood pressure variability (BPV) time series respectively. Heart rate variability (HRV) time series .
[0046] 120 circadian rhythm synchronization analysis module receives and The sequence. Its calculation process is similar to Example 1, but the preset frequency band for analysis is adjusted to the low-frequency (LF) band, exemplarily 0.04 Hz to 0.15 Hz. This band was selected because pressure receptor reflection activity is primarily manifested within this frequency range. The module calculates the amplitude-squared coherence function within this frequency band. The mean of the pressure receptors is multiplied by a scaling factor (e.g., 100) to generate PSPT values based on pressure receptor reflections.
[0047] For example, suppose a 64-second data window of BPV and HRV sequences is acquired. After analysis, the average coherence value within the [0.04, 0.15] Hz band is calculated to be 0.68. The resulting PSPT value is 68.0. This value will also be sent to the risk assessment and decision support module 140 and compared with the RIPE value of the selected rehabilitation intervention. The subsequent decision-making process is exactly the same as in Example 1.
[0048] By providing this embodiment, this application clearly teaches how to apply the core technical idea (i.e., to assess overall stability by quantifying the coupling between physiological subsystems) to different physiological signal pairs.
[0049] Method Implementation Examples
[0050] Reference Figure 2 This embodiment provides an ICU rehabilitation decision support method based on causal inference, which can be executed by the hardware and software modules in the above system embodiment. The steps of the method will be described in detail below.
[0051] S100: Acquire at least two real-time physiological time-series signals from the patient.
[0052] This step corresponds to the function of the data acquisition and preprocessing module 110 in the system embodiment. The execution of the method begins with the continuous acquisition of data from the bedside monitoring device. In a specific application, this step includes several sub-steps.
[0053] S110: Synchronously acquire raw physiological waveforms.
[0054] The system, through its interface with the monitoring equipment, simultaneously begins acquiring at least two signals at a preset, sufficiently high sampling frequency. For example, the electrocardiogram (ECG) signal is sampled at 500 Hz, and the respiratory waveform at 100 Hz. Time synchronization is crucial to ensure that the two sequences are temporally corresponding in subsequent analysis. Any discrepancy in timestamps can lead to incorrect estimations of the coupling relationship between them.
[0055] S120: Extract rhythmic feature sequences from the original waveform.
[0056] For the acquired ECG signals, a validated QRS wave detection algorithm is applied. This algorithm reliably identifies the main wave of each heartbeat and records the precise time point of its peak occurrence. By calculating the time difference between adjacent R wave peaks, an RR interval sequence in milliseconds is generated. This sequence reflects the beat-by-beat changes in the heart rhythm. For the respiratory waveform, the start and end points of each inspiration and expiration are determined by finding the local maxima and minima of the signal, thereby calculating the duration of each respiratory cycle and forming a respiratory cycle length sequence. Both sequences are non-equally spaced event sequences.
[0057] S130: Resample the rhythmic feature sequence into a time series with equal time intervals.
[0058] To use standard frequency domain analysis tools, such as Fourier transform, the input signal must be sampled at uniform time intervals. Therefore, the non-uniformly spaced sequence generated in S120 needs to be interpolated and resampled. A common method is to use cubic spline interpolation, which generates a smooth, continuous curve while preserving the dynamic characteristics of the original sequence. Then, sampling is performed on this curve at a fixed frequency. For example, 4 Hz is chosen as the resampling frequency, a frequency that covers the majority of the major frequency components of physiological rhythm signals. After this step, we obtain the heart rate variability (HRV) time series and respiratory rate variability (RRV) time series, which now have the same time axis and sampling rate.
[0059] S140: Perform data window segmentation on the time series.
[0060] The continuous HRV and RRV data streams are segmented into fixed-length, potentially overlapping, data windows. The length of the data window requires a trade-off between frequency resolution and temporal resolution. A longer data window provides better frequency resolution but reduces the response time to rapid changes in physiological state. Conversely, a shorter data window offers better real-time performance but poorer frequency resolution. For example, a 64-second data window is chosen, corresponding to 256 data points at a 4 Hz sampling rate. This 64-second window length represents a technical balance between ensuring at least six complete respiratory cycles within a single window for stable spectral estimation (the frequency resolution requirement) and maintaining a rapid response to changes in the patient's physiological state (the temporal resolution requirement). A 50% overlap rate can be chosen, meaning the results are calculated and updated every 32 seconds to improve real-time performance. These segmented data windows serve as the basic units for subsequent calculations.
[0061] S200: Based on the transspectral coherence between the at least two physiological time-series signals, determine a perturbation tolerance index characterizing the dynamic stability of the patient's physiological system.
[0062] This step is the core of the method and corresponds to the function of the physiological rhythm synchronicity analysis module 120. It transforms the input time-domain signal into a single, physiologically significant stability index, PSPT.
[0063] For example, the internal calculation process for this step is as follows: S210: Apply a window function to each data window.
[0064] To suppress the Gibbs phenomenon and spectral leakage caused by the Fourier transform, a window function, such as the Hanning window or the Blackman window, needs to be applied to each 256-point data window. This allows the values at both ends of the data window to decay smoothly to zero, reducing the abrupt changes caused by data truncation.
[0065] S220: Perform Fourier transform.
[0066] Perform a Fast Fourier Transform (FFT) on the windowed HRV and RRV sequences to transform the signals from the time domain to the frequency domain. The output is the amplitude and phase information of each signal at different frequency components.
[0067] S230: Calculate the power spectral density and cross-power spectral density.
[0068] Based on the FFT results, the power spectral density (PSD) of each signal and the cross-power spectral density (CPSD) between the two signals are calculated. The PSD describes the distribution of signal power across frequencies, while the CPSD describes the common power of the two signals across frequencies. To obtain a more robust estimate, the Welch method is typically used, which involves further subdividing the entire data window into smaller, overlapping segments, calculating the spectrum of each segment, and averaging the results.
[0069] S240: Calculate the amplitude squared coherence function.
[0070] Coherence is a normalized version of CPSD, measuring the degree of linear correlation between the output of one signal and the input of another signal at a given frequency. Its calculation formula is as follows: Coherence function The range of the function is [0, 1], where 1 indicates that the two signals are completely linearly correlated at frequency f, and 0 indicates that they are completely uncorrelated. This function is key to quantifying the synchronicity of two physiological rhythms.
[0071] S250: Integrate or calculate the mean within a specific frequency band.
[0072] Physiological studies have shown that the modulation of heart rate by respiration (respiratory sinus arrhythmia, RSA) is the main manifestation of cardiopulmonary coupling, and this phenomenon mainly occurs in the high-frequency band (HF), typically 0.15 Hz to 0.4 Hz in adults. This band was chosen because it precisely corresponds to the physiologically recognized high-frequency power spectrum range reflecting respiratory sinus arrhythmia in healthy adults, ensuring the physiological significance of the analysis. Therefore, this method focuses only on coherence within this band. By calculating the mean of the coherence function within this band, a single index representing the average strength of cardiopulmonary coupling can be obtained. For example, if the coherence values at 10 frequency points within this frequency band are calculated, then It's the arithmetic mean of these 10 values.
[0073] S260: Generate the Perturbation Tolerance Index (PSPT).
[0074] To make the indicators more intuitive, the average coherence value obtained in the previous step will be used. Perform a linear transformation, such as multiplying by 100. The resulting PSPT value ranges from 0 to 100. The higher the value, the stronger the cardiopulmonary coupling, the more stable the physiological system, and the higher the tolerance to external disturbances.
[0075] S300: Obtain the preset perturbation equivalent value corresponding to a rehabilitation intervention to be performed.
[0076] This step corresponds to the function of the rehabilitation intervention database module 130. In the clinical workflow, therapists first make selections on the system's user interface before planning a rehabilitation activity.
[0077] S310: Receive rehabilitation interventions selected by the user.
[0078] The system interface will provide a list containing all predefined rehabilitation activities, such as "turning over in bed," "sitting up," and "passive joint movement." The therapist clicks to select one of them.
[0079] S320: Query the database to retrieve the corresponding disturbance equivalent value (RIPE).
[0080] Based on the user's selection, the system uses the name of the rehabilitation activity as an index to look up the corresponding RIPE value in its internal database or configuration file. The calibration of RIPE values is an offline but systematic process, as described in the system embodiment above, ensuring objectivity and repeatability by combining the analytic hierarchy process (AHP) and expert scoring. This database is scalable and configurable, allowing hospitals to customize it according to their clinical practices.
[0081] S400: When the disturbance tolerance index is greater than the disturbance equivalent value, generate decision support information suggesting the implementation of the rehabilitation intervention to be performed.
[0082] This is the final step of the method, corresponding to the function of the risk assessment and decision support module 140, which transforms the calculation results into information that has direct clinical guidance.
[0083] S410: Compare the real-time PSPT value with the retrieved RIPE value.
[0084] This step performs a simple numerical comparison. For example, the system calculated a PSPT value of 85.0 in the most recent data window. The RIPE value corresponding to the rehabilitation activity "sitting upright at the bedside for 5 minutes" selected by the therapist is 90.0.
[0085] S420: Generate decision recommendations based on the comparison results.
[0086] In this example, because The comparison result is false. Therefore, the system will trigger a "high-risk" decision path. If the calculated PSPT value is 95.0, then... If the comparison result is true, the system will trigger the "low-risk" decision path.
[0087] S430: Outputs decision support information on the user interface.
[0088] Decision recommendations are presented to users through clear visual elements. For "low-risk" situations, a green icon and the text "Recommended to proceed" are displayed. For "high-risk" situations, a red icon and the text "High risk, postpone recommended" are displayed, possibly accompanied by an audio prompt. This intuitive feedback allows therapists to understand the current risk level at a glance, enabling them to make safer and more informed clinical decisions.
[0089] Those skilled in the art will understand that the above embodiments of this application can be implemented, in whole or in part, through software plus necessary general-purpose hardware platforms. In many cases, hardware implementation is preferable to software implementation. Based on this understanding, the technical solution of this application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the embodiments described in this application.
[0090] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
[0091] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.
[0092] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. An ICU rehabilitation decision support system based on causal inference, characterized in that, include: The data acquisition and preprocessing module is configured to acquire at least two real-time physiological time-series signals from the patient; The physiological rhythm synchronization analysis module is configured to determine a perturbation tolerance index characterizing the dynamic stability of the patient's physiological system based on the transspectral coherence between the at least two physiological time series signals. The rehabilitation intervention database module is configured to store multiple preset perturbation equivalent values corresponding to various rehabilitation intervention measures, and in response to a rehabilitation intervention selection instruction, provide the perturbation equivalent value corresponding to the rehabilitation intervention measure to be executed as indicated by the instruction. as well as The risk assessment and decision support module is configured to generate decision support information suggesting the implementation of the rehabilitation intervention measures to be performed when the disturbance tolerance index is greater than the disturbance equivalent value.
2. The system according to claim 1, characterized in that, The at least two physiological time series signals include a heart rate variability time series and a respiratory rate variability time series.
3. The system according to claim 2, characterized in that, The data acquisition and preprocessing module is specifically configured as follows: Simultaneous acquisition of electrocardiogram and respiratory signals; Based on the R-wave interval of the electrocardiogram signal, the heart rate variability time series is generated; and The respiratory rate variability time series is generated based on the respiratory cycle of the respiratory signal.
4. The system according to claim 1, characterized in that, The physiological rhythm synchronicity analysis module is specifically configured as follows: Spectral analysis is performed on the at least two physiological time series signals to obtain their respective power spectral density estimates and mutual power spectral density estimates. Based on the power spectral density estimation and the cross-power spectral density estimation, a squared amplitude coherence function within a preset frequency band is calculated. as well as The disturbance tolerance index is determined based on the integral or mean of the amplitude squared coherence function within the preset frequency band.
5. The system according to claim 4, characterized in that, The preset frequency band is 0.15 Hz to 0.4 Hz.
6. The system according to claim 1, characterized in that, The risk assessment and decision support module is also configured as follows: When the disturbance tolerance index is less than or equal to the disturbance equivalent value, decision support information is generated to postpone the implementation of the rehabilitation intervention measures to be implemented.
7. A causal inference-based ICU rehabilitation decision support method, characterized in that, include: Acquire at least two real-time physiological time-series signals from the patient; Based on the transspectral coherence between the at least two physiological time series signals, a disturbance tolerance index characterizing the dynamic stability of the patient's physiological system is determined. Obtain the preset perturbation equivalent value corresponding to a rehabilitation intervention to be implemented; as well as When the disturbance tolerance index is greater than the disturbance equivalent value, decision support information is generated to recommend the implementation of the rehabilitation intervention measures to be implemented.
8. The method according to claim 7, characterized in that, The at least two physiological time series signals include a heart rate variability time series and a respiratory rate variability time series.
9. The method according to claim 7, characterized in that, The step of determining a perturbation tolerance index characterizing the dynamic stability of a patient's physiological system based on the transspectral coherence between the at least two physiological time-series signals includes: Spectral analysis is performed on the at least two physiological time series signals to obtain their respective power spectral density estimates and mutual power spectral density estimates. Based on the power spectral density estimation and the cross-power spectral density estimation, a squared amplitude coherence function within a preset frequency band is calculated; and The disturbance tolerance index is determined based on the integral or mean of the amplitude squared coherence function within the preset frequency band.
10. The method according to claim 9, characterized in that, The preset frequency band is 0.15 Hz to 0.4 Hz.