Postoperative patient supervision system for neural interventional therapy

Abstract

The invention provides a postoperative patient supervision system for neural interventional therapy, and relates to the technical field of medical big data and artificial intelligence, and the system comprises a multi-modal data collection center, an interference feature decoupling unit, a trust capital quantification unit, a game strategy arbitration unit and a supervision execution unit. The multi-modal data acquisition center is configured to call time sequence monitoring data of a monitoring object; the trust capital quantification unit performs trust loss evaluation analysis on historical interaction feedback data; and the game strategy arbitration unit is configured to perform comparative analysis on the current clinical trust capital index and a preset trust threshold. According to the system, time-frequency domain matching is carried out on the residual error sequence and behavior state marking data, it is ensured that the system only carries out risk evolution prediction on neurogenic hemodynamic changes, and therefore the false alarm rate caused by external interference in a complex postoperative monitoring environment is remarkably reduced, and pure pathological feature components are extracted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical big data and artificial intelligence, and particularly relates to a postoperative patient monitoring system for neurointerventional therapy. BACKGROUND

[0002] With the wide application of neurointerventional therapy technology, the postoperative patient monitoring environment is becoming increasingly complex, and often faces multi-source noise interference caused by non-pathological factors such as pain, anxiety and body movement; such complexity brings severe challenges to the false alarm control ability and human-machine trust maintenance of the monitoring system; At present, the existing postoperative monitoring technology usually relies on the absolute threshold of physiological parameters for alarm triggering, or only removes noise based on a single signal filtering algorithm; such traditional method mainly focuses on the analysis of the physical characteristics of physiological signals themselves, and lacks effective monitoring of the interactive behavior and psychological trust state of medical staff; However, this single-dimensional monitoring strategy has significant limitations; when encountering high-frequency behavior-coupled noise component interference, the system often frequently outputs false alarms in order to ensure safety, resulting in serious alarm fatigue and response delay of medical staff; such mechanical alarm mechanism not only easily causes trust crisis, but also leads to the gradual abandonment of the system in long-term invalid interaction, and cannot dynamically balance between alarm sensitivity and user trust.

[0003] Therefore, how to accurately remove behavior noise in a complex interference environment and adaptively adjust the monitoring strategy according to the trust state of medical staff to prevent trust collapse has become a problem to be solved in the field. SUMMARY

[0004] In view of the deficiencies of the prior art, the purpose of the embodiments of the present application is to provide a postoperative patient monitoring system for neurointerventional therapy, which can solve the technical problems existing in the prior art.

[0005] A postoperative patient monitoring system for neurointerventional therapy, comprising a multi-modal data acquisition center, an interference feature decoupling unit, a trust capital quantification unit, a game strategy arbitration unit and a monitoring execution unit; The multi-modal data acquisition center is configured to retrieve time series monitoring data of a monitoring object, and send the time series monitoring data to the interference feature decoupling unit for signal source separation analysis, to obtain a pathological feature component and a behavior-coupled noise component, perform risk evolution prediction on the pathological feature component, and obtain an original risk prediction value; The trust capital quantification unit is configured to obtain historical interaction feedback data of the monitoring execution unit, perform trust loss evaluation analysis on the historical interaction feedback data, and calculate a current clinical trust capital index representing the reliability of the system; The game strategy arbitration unit is configured to compare and analyze the current clinical trust capital index with a preset trust threshold value: if the current clinical trust capital index is greater than or equal to the preset trust threshold value, a proactive supervision signal is generated; if the current clinical trust capital index is less than the preset trust threshold value, a defensive supervision signal is generated; In response to generating the proactive supervision signal, the supervision execution unit performs a high-sensitivity alarm operation based on the original risk prediction value; In response to generating the defensive supervision signal, the supervision execution unit is configured to invoke a preset explainability verification protocol, perform a degradation check process on the original risk prediction value, generate a modified risk output value, and perform an adaptive intervention operation based on the modified risk output value.

[0006] The signal source separation analysis process is as follows: The acquisition time period of the time series monitoring data is acquired, and the time series monitoring data is mapped into a preset generative adversarial network model to generate a digital twin simulation waveform corresponding to the acquisition time period, and a residual sequence between the time series monitoring data and the digital twin simulation waveform is calculated; The behavior state marker data of the monitoring object is obtained, the residual sequence is matched with the behavior state marker data in time-frequency domain correlation, if the matching is successful, the high-frequency fluctuation part in the residual sequence is set as the behavior coupling noise component, and the value obtained by subtracting the behavior coupling noise component from the time series monitoring data is set as the pathological feature component.

[0007] The trust loss evaluation analysis process is as follows: The latest alarm response delay duration in the historical interaction feedback data is acquired, and a preset effective rescue time window threshold value is acquired, and the latest alarm response delay duration is compared and analyzed with the effective rescue time window threshold value; If the latest alarm response delay duration is less than the effective rescue time window threshold value, a preset trust maintenance coefficient is selected; if the latest alarm response delay duration is greater than or equal to the effective rescue time window threshold value, a preset trust decay coefficient is selected, wherein the value of the trust decay coefficient is greater than the value of the trust maintenance coefficient; The total number of historical false alarms in the historical interaction feedback data is acquired, the product of the total number of historical false alarms and the currently selected coefficient is calculated to obtain a trust penalty value; the difference between a preset initial trust reference value and the trust penalty value is calculated, and the difference is set as the current clinical trust capital index.

[0008] The degradation check process is as follows: calculating a difference between the preset trust threshold and the current clinical trust capital index, determining a corresponding trust deficit level based on the difference; calling a corresponding feature filtering window parameter from a preset strategy library according to the trust deficit level, wherein the higher the trust deficit level is, the longer the time length of the corresponding feature filtering window parameter is; wherein the preset strategy library is a mapping table storing a functional mapping relationship between trust deficit levels and feature filtering window parameters, which pre-stores discrete trust deficit levels and corresponding numerical feature filtering window parameters, for realizing parameter retrieval and matching; performing time smoothing processing on the original risk prediction value by using the feature filtering window parameter, filtering out transient peaks in the original risk prediction value with a duration less than the feature filtering window parameter, to obtain smoothed risk trend data; setting the smoothed risk trend data as the corrected risk output value, wherein the alarm triggering sensitivity of the corrected risk output value is lower than that of the original risk prediction value.

[0009] When the aggressive supervision signal is generated, the game strategy arbitration unit is further configured to perform the following operations: obtaining a change slope of the pathological feature component, and comparing the change slope with a preset precursor feature slope threshold; If the change slope is greater than or equal to the precursor feature slope threshold, the original risk prediction value is directly marked as a first critical state, and the supervision execution unit is instructed to output an alarm instruction forcibly under the condition of ignoring the behavior coupling noise component.

[0010] The behavior coupling noise component represents a postoperative sympathetic storm artifact, and the feature extraction process includes: extracting blood pressure fluctuation amplitude and heart rate variability in the time series monitoring data to construct a multi-dimensional physiological feature vector; inputting the multi-dimensional physiological feature vector into a preset artifact recognition classifier, and if the classification result output by the artifact recognition classifier indicates non-neurogenic fluctuation, it is determined that the waveform segment corresponding to the multi-dimensional physiological feature vector is the behavior coupling noise component.

[0011] The adaptive intervention operation process is as follows: When the corrected risk output value exceeds a preset safety baseline, the supervision execution unit generates a verification request signal containing evidence traceability information; The verification request signal is used to simultaneously display the elimination criteria map of the behavior coupling noise component while outputting an alarm prompt, record the user's confirmation operation of the verification request signal, and feed the confirmation operation back to the trust capital quantification unit to update the historical interaction feedback data.

[0012] The trust capital quantification unit is also used to perform the following prediction operations: Extract multiple current clinical trust capital indices from historical time periods and construct a trust depletion prediction curve using a time series regression algorithm; Calculate the remaining number of alarms required for the value of the trust depletion prediction curve to drop to the preset response failure zero point, and set the remaining number of alarms as the security game budget value; The secure game budget value is compared with a preset budget security threshold: if the secure game budget value is less than or equal to the budget security threshold, a trust warning signal is generated and sent to the game strategy arbitration unit; The game strategy arbitration unit increases the preset trust threshold in response to receiving the trust warning signal.

[0013] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: Firstly, this system utilizes a multimodal data acquisition and interference feature decoupling unit to construct a digital twin simulation waveform of each patient using an improved generative adversarial network, which serves as a physiological benchmark for calculating residual sequences. The system performs time-frequency domain matching between the residual sequences and behavioral state labeling data, accurately identifying and eliminating behaviorally coupled noise components caused by non-pathological factors such as pain, anxiety, or body movement. This mechanism ensures that the system only predicts the risk evolution of neurogenic hemodynamic changes, thereby significantly reducing the false alarm rate caused by external interference in complex postoperative monitoring environments and extracting pure pathological feature components.

[0014] Secondly, this system innovatively introduces a trust capital quantification unit, transforming the abstract psychological state of medical staff into a calculable clinical trust capital index. The system is not limited to the analysis of physiological signals, but dynamically assesses the level of trust medical staff have in the equipment by monitoring historical interaction feedback data, particularly alarm response delay duration and the total number of historical false alarms. This mechanism reveals the dynamic principle of response delay as a sign of trust collapse, enabling the system to keenly perceive a trust crisis before being completely abandoned by medical staff and adjust its strategy according to the trust depletion prediction curve, preventing the system from falling into irreversible trust bankruptcy.

[0015] Thirdly, the system realizes adaptive switching of the monitoring mode through the game strategy arbitration unit. In the bonus period with a high trust capital index, the system executes the aggressive monitoring mode, adopts high-sensitivity alarm to ensure zero false negatives; in the deficit period with a low trust capital index, the system automatically switches to the defensive monitoring mode, calls the explainable verification protocol and uses the feature filtering window for time smoothing processing to filter out transient peaks. This tactical retreat strategy trades off part of the time resolution for extremely high alarm specificity, effectively reducing the nuisance alarm, and helps to gradually repair the human-machine trust relationship when the medical staff lack confidence, and maintains the long-term on-orbit operation of the system.

[0016] Fourthly, the system balances safety and transparency in extreme cases while pursuing low false positives; on the one hand, the system is provided with a forced alarm fusing mechanism, when the change slope of the pathological feature component is detected to exceed the premonitory feature slope threshold, the system will determine it as a first-class critical state, directly ignore any noise interference and force output alarm to ensure that catastrophic lesions are not missed; on the other hand, the system can generate a verification request signal containing evidence trace information in the defensive mode, which directly shows the basis atlas for noise elimination to the medical staff, and this visual evidence chain greatly enhances the persuasiveness of the alarm, assisting doctors to make clinical decisions quickly. BRIEF DESCRIPTION OF DRAWINGS

[0017] The application will be further explained below with reference to the drawings and embodiments: Figure 1 is a system block diagram of a postoperative patient monitoring system for neurointerventional therapy provided by an embodiment of the application. DETAILED DESCRIPTION

[0018] In order for those skilled in the art to better understand the technical solutions in the embodiments of the present application, the technical solutions of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. It should be understood that these descriptions are exemplary and are not intended to limit the scope of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0019] The exemplary embodiments will be described in detail here, and their examples are shown in the drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementation described in the following exemplary embodiments does not represent all the implementations consistent with the present application.

[0020] Embodiment 1 Reference Figure 1As shown, a postoperative patient monitoring system for neurointerventional therapy includes a multimodal data acquisition center, an interference feature decoupling unit, a trust capital quantification unit, a game strategy arbitration unit, and a monitoring execution unit. The multimodal data acquisition center is configured to retrieve time-series monitoring data of the monitored object and send the time-series monitoring data to the interference feature decoupling unit for signal source separation analysis to obtain pathological feature components and behavioral coupling noise components. Risk evolution prediction is then performed on the pathological feature components to obtain the original risk prediction value. The trust capital quantification unit is configured to acquire historical interaction feedback data from the monitoring execution unit, perform trust loss assessment analysis on the historical interaction feedback data, and calculate the current clinical trust value characterizing the system's reliability. The clinical trust capital index is used as the basis for the game strategy arbitration unit to compare and analyze the current clinical trust capital index with a preset trust threshold. If the current clinical trust capital index is greater than or equal to the preset trust threshold, an aggressive regulatory signal is generated; if the current clinical trust capital index is less than the preset trust threshold, a defensive regulatory signal is generated. In response to the generation of the aggressive regulatory signal, the regulatory execution unit performs a high-sensitivity alarm operation based on the original risk prediction value. In response to the generation of the defensive regulatory signal, the regulatory execution unit is configured to call a preset interpretability verification protocol to perform downgrade verification processing on the original risk prediction value, generate a corrected risk output value, and perform an adaptive intervention operation based on the corrected risk output value.

[0021] This embodiment details the overall architecture and data flow logic of a post-neurointerventional therapy monitoring system, aiming to construct an intelligent monitoring closed loop with clinical metacognitive capabilities. First, the system establishes a physical connection through a multimodal data acquisition center, retrieving raw signal streams from bedside monitors and wearable devices in real time. This data stream encompasses high-frequency sampled cerebral blood flow velocity and systemic physiological parameters. Subsequently, an interference feature decoupling unit acts as the system's sensing filter, introducing a dynamic decoupling mechanism in the signal source separation analysis to decompose the mixed raw waveforms into pure pathological features and behavioral noise. Following this, the system targets the pathological feature components... Risk evolution prediction is performed. To capture the uncertainty and dynamic trends of risk evolution over time, the system employs an Ornstein-Uhlenbeck process framework for modeling, a framework commonly used to describe mean-reverting stochastic processes. In this application scenario, the risk state... Modeled as surrounding a pathological stimulus A stochastic process driven by a time-varying mean that recovers. The constructed risk state evolution equation is as follows:

[0022] in, This represents the original risk forecast value (dimensionless); is the decay rate constant (1 / s) characterizing the rate of natural decay of the risk state; is the excitation gain constant (unit depends on the specific physiological signal, e.g. 1 / s for blood pressure, 1 / s2 for heart rate, etc.), characterizing the strength of the driving force of the risk state change; is the pathological feature component (unit depends on the specific physiological signal, e.g. mmHg for blood pressure, cm / s for cerebral blood flow velocity, etc.), if is the blood pressure, then the unit is 1 / (s mmHg)), characterizing the unit pathological feature component of the risk state change; is the noise intensity constant (1 / √s), reflecting the random fluctuations not captured by the model; is the pathological feature component (unit depends on the specific physiological signal, e.g. mmHg for blood pressure, cm / s for cerebral blood flow velocity, etc.); denotes the increment of the standard Wiener process, which follows the distribution at time step .

[0023] The estimation of parameters , , is based on the historical case data. The system discretizes the above SDE using Euler-Maruyama discretization method, and constructs the likelihood function based on the transition probability of the discretized sequence, to obtain the optimal parameters by maximum likelihood estimation (MLE) on the synchronized sequence of and .

[0024] The system solves the stochastic differential equation using Euler-Maruyama discretization method, sets the time step , and iteratively calculates the value of at the current time; wherein the parameters and are obtained by constructing the likelihood function based on the distribution properties of , and through maximum likelihood estimation on the historical case data; Specifically, based on the Markov property after discretization, the system constructs the negative log-likelihood function of the parameters as follows:

[0025] L denotes the loss function, denotes the total number of samples, denotes the conditional mean.

[0026] wherein the conditional mean is defined as ; by minimizing the objective function ​The system can reverse engineer the optimal dynamic parameters from noisy historical data. Based on this, the trust capital quantification unit works in parallel; it does not directly process physiological signals but instead monitors the feedback logs of the human-computer interaction interface in real time, quantifying the degree of psychological dependence of medical staff on the system, i.e., the current clinical trust capital index. Next, the game strategy arbitration unit, based on... With preset trust threshold Based on the comparison results, the system's operating strategy is dynamically adjusted: when When the system determines that it is in a period of high trust, it activates an aggressive regulatory mode to pursue zero false negatives; conversely, when... When the system determines that it is in a period of trust deficit, it automatically switches to defensive monitoring mode, sacrificing some real-time performance in exchange for high confidence by calling the interpretability verification protocol. Finally, the monitoring execution unit executes high-sensitivity alarms or adaptive intervention operations with evidence chain display according to the arbitration instructions, thereby completing a complete closed loop from signal perception to decision execution. This represents the discretization time step.

[0027] Example 2 The signal source separation and analysis process is as follows: The acquisition period of the time-series monitoring data is collected, and the time-series monitoring data is mapped to a preset generative adversarial network model to generate a digital twin analog waveform corresponding to the acquisition period. The residual sequence between the time-series monitoring data and the digital twin analog waveform is calculated. Behavioral state labeling data of the monitored object is obtained, and the residual sequence is matched with the behavioral state labeling data in the time-frequency domain. If the match is successful, the high-frequency fluctuation portion in the residual sequence is extracted and set as the behavioral coupling noise component, and the value obtained by subtracting the behavioral coupling noise component from the time-series monitoring data is set as the pathological feature component.

[0028] This embodiment details the core signal source separation and analysis logic in the interference feature decoupling unit. This logic uses an improved generative adversarial network to construct an individualized physiological benchmark. First, the system extracts the current acquisition time period. Time series monitoring data within This is then mapped to a predefined model. To address the issues of standard GANs failing to generate waveforms corresponding to specific inputs and the tendency for latent vectors to wander outside the effective manifold, this model constructs a variational encoder-decoder adversarial architecture. This architecture includes an encoder... Generator and discriminator The system constructs a hybrid objective function that includes reconstruction error, KL divergence regularization, and bidirectional adversarial constraints, as follows:

[0029] In the formula, Represents the distribution of real data Find the expected value; Indicates the prior distribution Hidden variables in mid-sample Find the expected value; Represents the natural logarithm function. Represents a mixed objective function. Indicates the reconstructed weights. Indicates the regularization weight. divergence, This represents the prior distribution. The output distribution of the constraint encoder approximates a standard normal distribution. ,and The waveform generated by the explicit constraint reconstructing the path must possess realistic statistical characteristics; based on this model, the system executes the deduction steps: firstly, through... The observed data is encoded into latent feature vectors, and then... Generate digital twin analog waveforms corresponding to the acquisition period. This waveform represents the ideal state of the patient under current physiological benchmarks without external interference, and retains high-frequency details. Next, the difference between the observed data and the digital twin simulated waveform is calculated, yielding the residual sequence formula as follows:

[0030] in, Represents the residual sequence. This represents time-series monitoring data. This represents the analog waveform of a digital twin.

[0031] The sequence physically includes all anomalous fluctuations deviating from the baseline; subsequently, external behavioral state marker data from accelerometers or video analytics are introduced. ,Will and Perform cross-correlation analysis in the time and frequency domains; calculate the residual sequence. With acceleration data Maximum cross-correlation number ,like If the power spectral density overlap rate between the two components in the 10Hz-20Hz frequency band is greater than 60%, they are considered a match. The system determines that this fluctuation originates from non-pathological factors and extracts it as a behavioral coupling noise component. Finally, the subtraction operation is performed as follows:

[0032] in, Indicates pathological feature components, This represents the behavioral coupled noise component.

[0033] Thus, a pure pathological component is obtained which only retains the neurogenic hemodynamic changes.

[0034] Embodiment 3 The trust loss evaluation analysis process is as follows: the latest alarm response delay duration in the historical interaction feedback data is obtained, and a preset effective rescue time window threshold is obtained, and the latest alarm response delay duration is compared and analyzed with the effective rescue time window threshold; if the latest alarm response delay duration is less than the effective rescue time window threshold, a preset trust maintenance coefficient is selected; if the latest alarm response delay duration is greater than or equal to the effective rescue time window threshold, a preset trust decay coefficient is selected, wherein the value of the trust decay coefficient is greater than the value of the trust maintenance coefficient; the total number of historical false alarms in the historical interaction feedback data is obtained, the product of the total number of historical false alarms and the currently selected coefficient is calculated, and a trust penalty value is obtained; the difference between the preset initial trust reference value and the trust penalty value is calculated, and the difference is set as the current clinical trust capital index.

[0035] This embodiment details the mathematical modeling process of the trust capital quantization unit, aiming to convert the abstract medical psychological state into a calculable numerical index; first, the system extracts the key time sequence parameter, i.e. the latest alarm response delay duration from the interaction log, and compares it with the preset effective rescue time window threshold ; the setting of the effective rescue time window threshold is based on the survival curve of cerebral ischemic penumbra, and is set as the statistical time critical value of irreversible nerve damage , i.e. minutes, minus the emergency preparation time ; then, the instantaneous loss factor is determined based on the comparison result, and the logic is as follows: if

[0036] wherein, represents the instantaneous loss factor, represents the response delay duration, represents the effective rescue time window threshold, wherein: is set to 0.05, which is based on the average cognitive wear statistical value of normal medical operation; is set to 0.8, and satisfies , which is derived from the weighted statistics of overtime response in historical medical dispute data; , , The setting is based on the questionnaire survey and behavior log analysis of 50 medical staff.

[0037] To avoid the nonlinear collapse of the trust capital index caused by a single interaction, the system introduces an inertia smoothing mechanism to calculate the current selected coefficient , as follows:

[0038] wherein, represents the current selected coefficient, represents the memory weight, represents the coefficient at the previous time.

[0039] wherein is set to , ensuring that the trust evaluation has historical memory; then, combined with the number of false positives in the sliding time window , the trust penalty value generated by this interaction is calculated ; finally, the initial trust reference value is used to perform subtraction update; here, the initial trust reference value is preset to a dimensionless number , symbolizing the complete trust state of the medical staff to the device when the system is first started, i.e. 100%; the calculation formula is as follows:

[0040] wherein, represents the current clinical trust capital index, represents the initial trust reference value, represents the trust penalty value. This embodiment innovatively introduces response delay as a nonlinear weighting factor for trust evaluation, revealing the dynamic mechanism of clinical trust collapse; when the medical staff starts to delay processing alarms due to distrust, the system can sensitively capture the change in this behavior pattern, and gradually reduce the trust capital index through the smoothly increasing penalty value ; this mechanism simulates the vicious cycle in the real world where suspicion leads to delay, and delay accelerates failure, enabling the system to perceive the crisis before being completely abandoned and providing a quantitative basis for subsequent strategy adjustment.

[0041] After calculating the difference, the system normalizes the current clinical trust capital index to ensure that it is within a reasonable representation range, for example, limited to the interval : This processing simulates the actual situation where the trust capital is at least zero and at most the initial value .

[0042] Embodiment 4 The degradation check processing process is as follows: calculating the difference between the preset trust threshold and the current clinical trust capital index, determining the corresponding trust deficit level based on the difference; according to the trust deficit level, calling the corresponding feature filtering window parameter from the preset strategy library, wherein the higher the trust deficit level, the longer the time length of the corresponding feature filtering window parameter; wherein the preset strategy library is a mapping table storing the functional mapping relationship between the trust deficit level and the feature filtering window parameter, which pre-stores discrete trust deficit levels and corresponding numerical feature filtering window parameters, for realizing parameter retrieval and matching, the preset strategy library realizes the smoothness control under different trust degree scenes by establishing a quantitative mapping model between the trust deficit level and the feature filtering window parameter; using the feature filtering window parameter to perform time smoothing processing on the original risk prediction value, filtering out transient peaks in the original risk prediction value with a duration less than the feature filtering window parameter, to obtain smoothed risk trend data; setting the smoothed risk trend data as the corrected risk output value, wherein the alarm triggering sensitivity of the corrected risk output value is lower than that of the original risk prediction value.

[0043] This embodiment details the degradation check processing logic under the defense supervision mode, which sacrifices time resolution to obtain the specificity of the alarm; first, calculate the trust deficit , and quantize the trust deficit level accordingly, the process of quantizing the trust deficit level follows the principle of rounding down: , wherein represents the rounding down symbol. Then, query the preset strategy library with as the index to extract the corresponding feature filtering window parameter , which is positively correlated with the deficit level, that is, the lower the trust, the wider the window; the specific functional mapping relationship is defined as:

[0044] , wherein represents the feature filtering window parameter, represents the basic window length, represents the level expansion coefficient, represents the trust deficit level; the quantitative mapping model is realized by a linear function , which ensures that the feature filtering window parameter can be widened step by step with the increase of the trust deficit;

[0045] is set to seconds, is set to Seconds / level; subsequently, using Original risk forecast The moving average filtering operation is performed, and the specific calculation formula is as follows:

[0046] This is represented as the corrected risk output value. This is represented as the original risk forecast value. Represented as feature filtering window parameters, Indicates the current time. It is represented as a discrete lag index.

[0047] This formula aims to filter out those with a duration of less than [a certain value]. The transient spike signal; finally, the smoothed signal... The output is a corrected risk output value, which has a significantly lower alarm trigger sensitivity than the original value and only reflects long-term and stable risk trends.

[0048] In this embodiment, when the system encounters a crisis of trust, a slow but robust monitoring strategy is automatically activated. By dynamically extending the observation window, the system effectively filters out occasional physiological fluctuations and sensor noise. Although this delays the alarm time to some extent, it greatly improves the positive predictive value of the alarm. This tactical retreat strategy is particularly crucial in scenarios where medical staff lack confidence in the equipment. It gradually restores the trust relationship between humans and machines by reducing harassing alarms.

[0049] Example 5 When the aggressive regulatory signal is generated, the game strategy arbitration unit is also used to perform the following operations: obtain the slope of change of the pathological feature component, compare the slope of change with a preset warning feature slope threshold; if the slope of change is greater than or equal to the warning feature slope threshold, directly mark the original risk prediction value as a first-level critical state, and instruct the regulatory execution unit to forcibly output an alarm command while ignoring the behavioral coupling noise component.

[0050] This embodiment details the mandatory alarm circuit breaker mechanism under aggressive monitoring mode, designed to address catastrophic acute lesions; firstly, the system analyzes pathological feature components. Perform first-order differential operations and calculate its slope. This indicator characterizes the rate of disease progression; then, Compared with the preset precursor feature slope threshold Perform real-time comparison; respond to , the system determines that it is currently in the outbreak period of extreme danger, that is, the moment of aneurysm rupture, at which time any denoising algorithm may mistakenly delete the true signal; therefore, the system immediately bypasses the conventional game logic and marks the original risk prediction value as a first-level critical state; finally, the command supervision execution unit forcibly outputs the highest priority alarm instruction, completely ignoring the existence of the behavior coupling noise component in the process.

[0051] Embodiment 6 The behavior coupling noise component represents a postoperative sympathetic storm artifact, and the feature extraction process includes: extracting the blood pressure fluctuation amplitude and the heart rate variability in the time series monitoring data, constructing a multi-dimensional physiological feature vector; inputting the multi-dimensional physiological feature vector into a preset artifact recognition classifier, and if the classification result output by the artifact recognition classifier indicates non-neurogenic fluctuation, determining that the waveform segment corresponding to the multi-dimensional physiological feature vector is the behavior coupling noise component.

[0052] This embodiment details the feature extraction and recognition process for postoperative sympathetic storm artifacts; first, two key dimensions of features are extracted in parallel from the time series monitoring data: blood pressure fluctuation amplitude and heart rate variability , which are combined to construct a multi-dimensional physiological feature vector ; this vector comprehensively reflects the regulation state of the autonomic nervous system; then, is input to a pre-trained artifact recognition classifier, i.e., a support vector machine (SVM); this classifier is trained based on a set of expert-labeled sympathetic storm artifact samples , and uses a radial basis kernel function:

[0053] wherein, represents the kernel function, represents the input vector, represents the center vector, represents the kernel parameter.

[0054] wherein, the kernel parameter controls the distribution complexity after mapping the data to a high-dimensional feature space, and the penalty coefficient is used to balance the classification interval and the number of misclassified samples. In this embodiment, the system uses a combination of cross-validation and grid search to simultaneously optimize and , with the search range of set to , and the search range of , if the classification result output by the artifact identification classifier indicates that it is a non-neurogenic fluctuation, that is, caused by peripheral stimulation such as pain, anxiety, etc., the system determines that the current waveform segment belongs to the behavior-coupled noise component; finally, the waveform segment is marked for use by the subsequent decoupling unit.

[0055] Embodiment 7 The adaptive intervention operation process is as follows: when the correction risk output value exceeds the preset safety baseline, the supervision execution unit generates a verification request signal containing evidence traceability information; the verification request signal is used to output an alarm prompt while synchronously displaying the elimination basis atlas of the behavior-coupled noise component, and records the user's confirmation operation of the verification request signal, and feeds back the confirmation operation to the trust capital quantification unit to update the historical interaction feedback data.

[0056] This embodiment details the interaction logic in adaptive intervention operation, focusing on rebuilding user trust through white-box display; first, real-time monitoring of the correction risk output value , in response to the value exceeding the preset safety baseline , the system triggers a verification request signal; then, two layers of information are synchronously rendered on the user interface: one is the regular alarm prompt, and the other is the key evidence traceability information, that is, the elimination basis atlas of the behavior-coupled noise component; the atlas directly shows the noise segment identified by the system, that is, the body motion artifact, and the waveform comparison before and after it is eliminated; then, the system records the user's confirmation operation of the request, that is, confirming validity or marking false positives, completing a complete human-computer interaction; finally, the operation result is fed back to the trust capital quantification unit as a new data point, for real-time updating of the total number of historical false positives and response delay .

[0057] Embodiment 8 The trust capital quantification unit is also used to perform the following prediction operations: extracting a plurality of current clinical trust capital indices in a historical time period, and constructing a trust depletion prediction curve using a time series regression algorithm; calculating the remaining alarm number required for the value of the trust depletion prediction curve to drop to a preset response failure zero point, and setting the remaining alarm number as a safety game budget value; comparing the safety game budget value with a preset budget safety threshold value: if the safety game budget value is less than or equal to the budget safety threshold value, generating a trust warning signal and sending it to the game strategy arbitration unit; The game strategy arbitration unit increases the preset trust threshold in response to receiving the trust warning signal. The preset trust threshold is set based on a false alarm rate and a response delay statistical result of the first 100 patient interaction data.

[0058] The embodiment details the advanced prediction function of the trust capital quantification unit, aiming to prevent the system from falling into irreversible trust bankruptcy; first, extract a plurality of current clinical trust capital index sequences in a past time window ; then, linear fitting is performed on the sequence by using the least square method to construct a trust depletion prediction curve, the formula is as follows:

[0059] Among them, represents the prediction curve, represents the current index, represents the average depletion rate, represents the number of future alarms; the average depletion rate is calculated by the least square estimation formula, which clearly distinguishes the statistical characteristics of the time index and the capital value:

[0060] In the formula, is the average depletion rate; represents the sequence index in the historical time window (for example ), represents the arithmetic mean of the sequence index; represents the current clinical trust capital index value corresponding to the th time point, represents the arithmetic mean of all trust capital indexes in the historical window. Suppose there are trust capital indexes of time points in the historical window, denoted as sequence , and the corresponding time index is . is the arithmetic mean of , and is the arithmetic mean of .

[0061] Here is defined to avoid confusion with the index mean ; then, solve the equation to calculate the number of remaining alarms required for the curve to drop to the preset response failure zero point , and define this number as the safety game budget value ; represents the critical state of the doctor completely ignoring the alarm; finally, sent to the game strategy arbitration unit, in response to below the safety alert line, the arbitration unit will greatly increase the preset trust threshold , forcing the system to enter the defense mode in advance; the specific adjustment operation is as follows:

[0062] In the formula, is the adjusted threshold value, is the original threshold value, is the gain amplitude, is the sensitivity factor, is the safety game budget value, and the non-linear formula ensures that when the remaining budget tends to , the threshold value will be sharply increased, thereby tightening the monitoring strategy. In this embodiment, the gain amplitude is set to 0.5 to limit the maximum threshold value to not more than 50%; the sensitivity factor is set to 0.2, which is statistically fitted according to the average number of alarms abandoned by doctors in historical cases.

[0063] This embodiment gives the system a meta-cognitive ability of self-protection; by predicting the critical point of trust depletion, the system is no longer passively waiting to be abandoned by the user, but actively adjusts the strategy according to the remaining trust budget; on the brink of trust bankruptcy, the system can intelligently shrink the defense line to maintain the continuation of the system's vitality in long-term postoperative monitoring.

[0064] The present application covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the preferred embodiments of the present application, and the present application can also be fully understood without the description of these details to those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present application, and not to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent substitutions for part of the technical features; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. Any changes or substitutions easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered within the protection scope of the present application.

Claims

1. A post-operative patient monitoring system for neuro-interventional therapy, characterized by, The system comprises a multi-modal data acquisition center, an interference feature decoupling unit, a trust capital quantification unit, a game strategy arbitration unit, and a supervision execution unit. The multi-modal data acquisition center is configured to call time-series monitoring data of a monitoring object, and send the time-series monitoring data to the interference feature decoupling unit for signal source separation analysis to obtain a pathological feature component and a behavior coupling noise component, and perform risk evolution prediction on the pathological feature component to obtain an original risk prediction value. The trust capital quantification unit is configured to obtain historical interaction feedback data of the supervision execution unit, perform trust loss evaluation analysis on the historical interaction feedback data, and calculate a current clinical trust capital index representing the reliability of the system. The game strategy arbitration unit is configured to compare and analyze the current clinical trust capital index with a preset trust threshold value: if the current clinical trust capital index is greater than or equal to the preset trust threshold value, a proactive supervision signal is generated; if the current clinical trust capital index is less than the preset trust threshold value, a defensive supervision signal is generated. In response to generating the proactive supervision signal, the supervision execution unit performs a high-sensitivity alarm operation based on the original risk prediction value. In response to generating the defensive supervision signal, the supervision execution unit is configured to call a preset explainability verification protocol, perform degradation verification processing on the original risk prediction value, generate a modified risk output value, and perform an adaptive intervention operation based on the modified risk output value.

2. A post-operative patient monitoring system for neuro-interventional therapy according to claim 1, wherein, The signal source separation analysis process is as follows: The time period of the time-series monitoring data is collected, and the time-series monitoring data is mapped into a preset generative adversarial network model to generate a digital twin simulation waveform corresponding to the collection period, and the residual sequence between the time-series monitoring data and the digital twin simulation waveform is calculated. The behavior state marker data of the monitoring object is obtained, the residual sequence is matched with the behavior state marker data in the time-frequency domain, if the matching is successful, the high-frequency fluctuation part in the residual sequence is set as the behavior coupling noise component, and the value obtained by subtracting the behavior coupling noise component from the time-series monitoring data is set as the pathological feature component.

3. The post-operative patient monitoring system for neurological interventions of claim 1, wherein, The trust loss evaluation analysis process is as follows: The latest alarm response delay duration in the historical interaction feedback data is obtained, and a preset effective rescue time window threshold value is obtained, and the latest alarm response delay duration is compared and analyzed with the effective rescue time window threshold value. If the latest alarm response delay duration is less than the effective rescue time window threshold value, a preset trust maintenance coefficient is selected; if the latest alarm response delay duration is greater than or equal to the effective rescue time window threshold value, a preset trust decay coefficient is selected, wherein the value of the trust decay coefficient is greater than the value of the trust maintenance coefficient. The total number of historical false alarms in the historical interaction feedback data is obtained, the product of the total number of historical false alarms and the currently selected coefficient is calculated to obtain a trust penalty value. The difference between the preset initial trust benchmark value and the trust penalty value is calculated, and the difference is set as the current clinical trust capital index.

4. The post-operative patient monitoring system for neurological interventions of claim 1, wherein, The degradation check process is as follows: The difference between the preset trust threshold value and the current clinical trust capital index is calculated, and the corresponding trust deficit level is determined based on the difference; According to the trust deficit level, the corresponding feature filtering window parameter is called from the preset strategy library, wherein the higher the trust deficit level, the longer the time length of the corresponding feature filtering window parameter; wherein the preset strategy library is a mapping table storing the functional mapping relationship between the trust deficit level and the feature filtering window parameter, which pre-stores discrete trust deficit levels and corresponding numerical feature filtering window parameters, for realizing parameter retrieval and matching; The original risk prediction value is time-smoothed using the feature filtering window parameter to filter out transient peaks in the original risk prediction value with a duration less than the feature filtering window parameter, to obtain smoothed risk trend data; The smoothed risk trend data is set as the modified risk output value, wherein the alarm triggering sensitivity of the modified risk output value is lower than that of the original risk prediction value.

5. The post-operative patient monitoring system for neurological interventions of claim 1, wherein, When the aggressive supervision signal is generated, the game strategy arbitration unit is further configured to perform the following operations: Obtain the change slope of the pathological feature component, and compare the change slope with a preset precursor feature slope threshold value; If the change slope is greater than or equal to the precursor feature slope threshold value, the original risk prediction value is directly marked as a first critical state, and the supervision execution unit is instructed to output an alarm instruction forcibly while ignoring the behavior coupling noise component.

6. The post-operative patient monitoring system for neurological interventions of claim 2, wherein, The behavior coupling noise component represents a postoperative sympathetic storm artifact, and the feature extraction process includes: Extract the blood pressure fluctuation amplitude and heart rate variability in the time series monitoring data to construct a multi-dimensional physiological feature vector; Input the multi-dimensional physiological feature vector into a preset artifact recognition classifier, and if the classification result output by the artifact recognition classifier indicates a non-neurogenic fluctuation, it is determined that the waveform segment corresponding to the multi-dimensional physiological feature vector is the behavior coupling noise component.

7. The post-operative patient monitoring system for neurological interventions of claim 1, wherein, The adaptive intervention operation process is as follows: When the modified risk output value exceeds a preset safety baseline, the supervision execution unit generates a verification request signal containing evidence trace information; The verification request signal is used to output an alarm prompt while synchronously displaying a basis map for excluding the behavior coupling noise component, and records a user confirmation operation on the verification request signal, and feeds back the confirmation operation to the trust capital quantification unit to update the historical interaction feedback data.

8. The post-operative patient monitoring system for neurological interventions of claim 3, wherein, The trust capital quantification unit is further configured to perform the following prediction operations: Extract a plurality of current clinical trust capital indexes in a historical time period, and construct a trust depletion prediction curve using a time series regression algorithm; Calculate the remaining alarm number required for the numerical value of the trust depletion prediction curve to drop to a preset response failure zero point, and set the remaining alarm number as a safety game budget value; comparing the security game budget value with a preset budget security threshold value; if the security game budget value is less than or equal to the budget security threshold value, a trust warning signal is generated and sent to the game strategy arbitration unit; the game strategy arbitration unit, in response to receiving the trust warning signal, increases the preset trust threshold value.

Images