Postoperative care risk warning system for neurosurgical patients

Abstract

This invention discloses a postoperative care risk warning system for neurosurgical patients, belonging to the field of postoperative care risk warning technology. This system integrates multi-source postoperative data, including EEG, physiological indicators, behavior, and inflammation, through a data acquisition module. It uses a sub-risk assessment module to identify abnormalities in individual indicators, and a comprehensive risk assessment module to further integrate data for overall risk judgment when all individual indicators are normal. Finally, a warning module issues tiered warnings, achieving multi-dimensional data fusion analysis. This system can promptly detect abnormalities in individual indicators and uncover potential overall risks through a comprehensive model, improving the comprehensiveness, accuracy, and timeliness of warnings. It provides a scientific and precise basis for risk judgment in postoperative care, assisting medical staff in early intervention, improving patient prognosis, and solving the problem of single-source and lacking comprehensive judgment in current neurosurgical postoperative care risk assessments.

Description

Technical Field

[0001] This invention relates to the field of postoperative nursing risk warning technology, specifically a postoperative nursing risk warning system for neurosurgical patients. Background Technology

[0002] Neurosurgical diseases often involve complex surgical interventions of the brain or spinal cord, which can lead to various postoperative complications, including increased intracranial pressure, insufficient cerebral perfusion, and abnormal cerebral oxygen saturation. The rapid and complex changes in postoperative conditions make it difficult for traditional nursing methods (relying on human experience and static scoring mechanisms) to achieve precise monitoring and rapid response. This inadequacy may result in the failure to detect deterioration in a timely manner, increasing the risk of complications. Furthermore, the medical and nursing workload in neurosurgical wards is significant, comparable to that in other departments, and the neurosurgical intensive care unit requires an even greater emphasis on timeliness.

[0003] Chinese invention patent application CN119694597A discloses a neurosurgical nursing assessment method based on clinical early warning. By acquiring historical vital sign data of monitored patients, an isolated forest algorithm is used to detect abnormal data points. Local weighted regression is used to smooth the time series data and remove noisy data. Interpolation is used to fill in missing values. A sliding window method is used to segment the time series data and extract trend and fluctuation characteristics. For each type of vital sign data, a time fluctuation curve is constructed based on the fluctuation characteristics. The postoperative stable period is used as a critical point, and the fluctuation index of the time fluctuation curve before the critical point is calculated. Based on the fluctuation index, an abnormal signal is identified by comparing it with a threshold output by a machine learning model, significantly improving the accuracy of clinical early warning.

[0004] However, existing methods for assessing postoperative care risks in neurosurgical patients often focus on single-dimensional data and fail to integrate heterogeneous information from multiple sources such as EEG, physiology, and behavior, making it difficult to capture the risks associated with the indicators. Furthermore, they lack the ability to explore the correlation and synergistic effects of data, making it impossible to comprehensively depict the patient's overall recovery status and easily overlooking complex risks. Therefore, they suffer from problems of singular assessment and lack of comprehensive judgment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a postoperative care risk warning system for neurosurgical patients, which solves the problem of single-faceted and lacking comprehensive judgment in the current postoperative care risk assessment for neurosurgical patients.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a postoperative care risk warning system for neurosurgical patients, comprising a data acquisition module, a sub-risk assessment module, a comprehensive risk assessment module, and an early warning module. Specifically: the data acquisition module acquires various postoperative indicator data of neurosurgical patients, including electroencephalogram (EEG) signal data, physiological indicator data, behavioral data, and inflammation data; the sub-risk assessment module performs sub-risk assessments on the various indicator data to obtain various risk assessment results, including EEG assessment results, physiological indicator assessment results, behavioral assessment results, and inflammation assessment results, with each risk assessment result including abnormal and normal assessment results; the comprehensive risk assessment module, when all risk assessment results are normal, fuses the various indicator data to obtain a trained comprehensive risk assessment model, processes the fused indicator data based on the trained comprehensive risk assessment model, and obtains a comprehensive risk assessment result, including a comprehensive assessment result of abnormality and a comprehensive assessment result of normality; the early warning module acquires the individual risk assessment results and the comprehensive risk assessment result, and issues early warnings based on these results.

[0007] Furthermore, the data acquisition module includes an electroencephalogram (EEG) signal acquisition unit, a physiological indicator acquisition unit, a behavioral data acquisition unit, and an inflammation data acquisition unit. Specifically: the EEG signal acquisition unit connects to a high-density electroencephalogram (EEG) device to acquire initial EEG signals from multiple electrode sites across the entire brain, and processes these initial EEG signals based on independent component analysis and bandpass filtering to obtain EEG signal data; the physiological indicator acquisition unit acquires basic postoperative physiological indicators for neurosurgical patients, preprocesses these indicators, and then standardizes them to obtain physiological indicator data; the behavioral data acquisition unit acquires patient movement data using a camera, identifies the patient's movement data based on a target detection algorithm to determine patient behavior, and simultaneously acquires behavioral data using an inertial measurement sensor; the inflammation data acquisition unit acquires postoperative clinical examination results, extracts initial inflammation data, and standardizes the initial inflammation data to obtain inflammation data.

[0008] Furthermore, the EEG signal data includes EEG frequency parameters, EEG time-domain parameters, and EEG spatial parameters; physiological indicator data includes basic vital signs parameters and neurosurgical specific parameters; behavioral data includes acceleration and angular velocity during movement corresponding to each patient's behavior; and inflammatory data includes blood inflammatory parameters and inflammatory mediator parameters.

[0009] Furthermore, the sub-risk assessment module includes an EEG signal risk assessment unit, a physiological indicator risk assessment unit, a behavioral risk assessment unit, and an inflammation risk assessment unit. Among them, the EEG signal risk assessment unit is used to identify abnormalities in EEG signal data based on an autoencoder and output EEG assessment results.

[0010] The physiological indicator risk assessment unit is used to obtain the physiological indicator assessment coefficient based on the physiological indicator data, and compare the physiological indicator assessment coefficient with the physiological indicator assessment threshold stored in the database: if the physiological indicator assessment coefficient is greater than the physiological indicator assessment threshold stored in the database, the physiological indicator assessment result is output as abnormal; if the physiological indicator assessment coefficient is not greater than the physiological indicator assessment threshold stored in the database, the physiological indicator assessment result is output as normal.

[0011] The behavioral risk assessment unit is used to obtain behavioral assessment coefficients based on behavioral data and compare the behavioral assessment coefficients with the behavioral assessment thresholds stored in the database: if the behavioral assessment coefficient is greater than the behavioral assessment threshold stored in the database, the output behavioral assessment result is "abnormal"; if the behavioral assessment coefficient is not greater than the behavioral assessment threshold stored in the database, the output behavioral assessment result is "normal".

[0012] The inflammation risk assessment unit is used to obtain an inflammation assessment coefficient based on inflammation data and compare the inflammation assessment coefficient with the inflammation assessment threshold stored in the database: if the inflammation assessment coefficient is greater than the inflammation assessment threshold stored in the database, the output inflammation assessment result is abnormal; if the inflammation assessment coefficient is not greater than the inflammation assessment threshold stored in the database, the output inflammation assessment result is normal.

[0013] Furthermore, the process of anomaly identification in EEG signal data by the autoencoder is as follows: Obtaining periodically updated abnormal EEG signal data from the database; training the autoencoder based on the abnormal EEG signal data, whereby the encoder in the autoencoder compresses the input abnormal EEG signal data into a low-dimensional EEG signal feature vector, and the decoder in the autoencoder reconstructs the low-dimensional EEG signal feature vector back to the original dimension, obtaining reconstructed abnormal EEG signal data; calculating the mean square error based on the abnormal EEG signal data and the reconstructed abnormal EEG signal data, and using the backpropagation algorithm to propagate the value of the loss function back to the parameters of the encoder and decoder, adjusting the parameters using gradient descent to continuously reduce the mean square error; when the mean square error is less than the set mean square error threshold, training ends, and the trained encoder is output, where the parameters include weights and biases.

[0014] The EEG signal data is input into the trained encoder, and after encoding and decoding, the reconstructed EEG signal data to be verified is obtained. The mean square error to be verified is calculated based on the reconstructed EEG signal data and the EEG signal data. If the mean square error to be verified is greater than the set error threshold, the output EEG evaluation result is abnormal; if the mean square error to be verified is not greater than the set error threshold, the output EEG evaluation result is normal.

[0015] Furthermore, basic vital signs parameters include heart rate, blood pressure, respiratory rate, and body temperature; neurosurgical specific parameters include intracranial pressure and blood oxygen saturation; the process of obtaining physiological indicator evaluation coefficients based on physiological indicator data is as follows: Obtain the patient's basic pathological information, including age, gender, weight, height, and underlying disease information; obtain the basic case information of historical patients who have undergone the same neurosurgical procedure stored in the case database; analyze the similarity between the patient's basic pathological information and the basic case information of historical patients based on cosine similarity, and determine the basic case information of historical patients most similar to the patient's basic pathological information; obtain the corresponding postoperative physiological indicator records stored in the database based on the most similar historical patient case information, including daily postoperative physiological indicator data; perform difference processing between the physiological indicator data of the day and the corresponding postoperative physiological indicator data in the patient's postoperative physiological indicator records to obtain the physiological indicator evaluation coefficients.

[0016] Furthermore, the process of obtaining behavioral evaluation coefficients based on behavioral data is as follows: obtain postoperative behavioral data of patients undergoing similar neurosurgical procedures; standardize the behavioral data and postoperative behavioral data and then perform cosine similarity analysis to obtain behavioral evaluation coefficients.

[0017] Furthermore, blood inflammatory parameters include the levels of various white blood cells, C-reactive protein, and procalcitonin; inflammatory mediator parameters include the levels of interleukin-6 and tumor necrosis factor-α. The process of obtaining the inflammation assessment coefficient based on inflammatory data is as follows: Obtain normal human inflammatory data, including normal human blood inflammatory parameters and normal human inflammatory mediator parameters; obtain postoperative inflammatory data of patients undergoing similar neurosurgical procedures, including postoperative blood inflammatory parameters and postoperative inflammatory mediator parameters; obtain a first inflammatory assessment factor based on the normal human inflammatory data and the inflammatory data; obtain a second inflammatory assessment factor based on the postoperative inflammatory data and the inflammatory data of patients undergoing similar neurosurgical procedures; and obtain the inflammatory assessment coefficient by weighted summation of the first and second inflammatory assessment factors.

[0018] Furthermore, the comprehensive risk assessment model is a gradient boosting decision tree model. The process of obtaining the comprehensive risk assessment result based on the comprehensive risk assessment model is as follows: the data of each indicator are dedimensionalized and concatenated into a comprehensive indicator vector; the comprehensive indicator vector is input into the comprehensive risk assessment model to obtain the predicted value output by each decision tree; the predicted values ​​are accumulated according to the set weights to obtain the probability that the comprehensive indicator vector belongs to the normal range; if the probability is greater than the set normal probability threshold, the comprehensive risk assessment result is output as normal; if the probability is not greater than the set normal probability threshold, the comprehensive risk assessment result is output as abnormal.

[0019] Furthermore, the process of issuing early warnings based on the results of each risk assessment and the comprehensive risk assessment is as follows: count the number of abnormal assessment results obtained from the sub-risk assessments. If the number is no more than two, a Level 1 warning is issued; if the number is more than two, a Level 2 warning is issued. If an abnormal comprehensive assessment result is received, a Level 2 warning is issued directly.

[0020] The present invention has the following beneficial effects:

[0021] This postoperative risk warning system for neurosurgical patients integrates multi-source postoperative data, including EEG, physiological indicators, behavior, and inflammation, through a data acquisition module. It uses a sub-risk assessment module to identify abnormalities in individual indicators, and a comprehensive risk assessment module to further integrate data for overall risk judgment when all individual indicators are normal. Finally, a tiered warning module issues warnings, achieving multi-dimensional data fusion analysis. This system can promptly detect abnormalities in individual indicators and uncover potential overall risks through a comprehensive model, improving the comprehensiveness, accuracy, and timeliness of warnings. It provides a scientific and precise basis for risk assessment in postoperative care, assisting medical staff in early intervention, improving patient prognosis, and solving the problem of single-source and lack of comprehensive judgment in current postoperative risk assessments for neurosurgical patients.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 This is a flowchart of the postoperative care risk warning system for neurosurgical patients according to the present invention.

[0024] Figure 2 This invention provides a flowchart of the process for obtaining inflammation assessment coefficients from inflammation data in the postoperative care risk warning system for neurosurgical patients. Detailed Implementation

[0025] Please see Figure 1 The present invention provides a technical solution: a postoperative nursing risk warning system for neurosurgical patients, comprising a data acquisition module, a sub-risk assessment module, a comprehensive risk assessment module, and a warning module, wherein: the data acquisition module is used to acquire various postoperative indicator data of neurosurgical patients, including electroencephalogram signal data, physiological indicator data, behavioral data, and inflammation data.

[0026] The data acquisition module includes an EEG signal acquisition unit, a physiological indicator acquisition unit, a behavioral data acquisition unit, and an inflammation data acquisition unit. Specifically, the EEG signal acquisition unit connects to a high-density EEG device to acquire initial EEG signals from multiple electrode sites across the entire brain, such as the international 10-20 system standard sites, capturing EEG signals in real time at a high sampling frequency of ≥1000Hz. The initial EEG signals are processed using independent component analysis (separating and removing bioelectrical artifacts such as electrooculography and electromyography to avoid interfering with the authenticity of the EEG signals) and bandpass filtering to obtain EEG signal data. Bandpass filtering is designed for different frequency bands (alpha waves 8-13Hz, beta waves 14-30Hz, gamma waves 30-100Hz, delta waves 0.5-4Hz, and theta waves 4-8Hz) to extract effective signal components related to brain functions such as wakefulness, sleep, and cognition.

[0027] The physiological index acquisition unit is used to acquire the basic physiological indexes of neurosurgical patients after surgery. The basic physiological indexes are preprocessed (outliers are removed by the 3σ principle and missing data are filled by linear interpolation) and then standardized (by Z-score or Min-Max normalization) to obtain the physiological index data. The behavioral data acquisition unit is used to acquire patient action data based on a camera and to identify the patient action data based on a target detection algorithm (such as YOLOv5) to determine the patient's behavior (such as turning over, sitting up, getting out of bed, etc.). At the same time, behavioral data is acquired based on an inertial measurement unit (IMU, including accelerometer and gyroscope). The inflammation data acquisition unit is used to acquire the patient's postoperative clinical examination results, extract the initial inflammation data, and obtain the inflammation data after standardization of the initial inflammation data.

[0028] EEG signal data includes EEG frequency parameters (energy percentage of each frequency band, power spectral density), EEG time-domain parameters (amplitude, latency), and EEG spatial parameters (phase difference between electrodes, correlation); physiological index data includes basic vital signs parameters and neurosurgical specific parameters; behavioral data includes acceleration and angular velocity during movement corresponding to each patient's behavior; inflammatory data includes blood inflammatory parameters and inflammatory mediator parameters.

[0029] Electroencephalogram (EEG) signals directly reflect the state of neurological function (such as epileptiform discharges and ischemic slow waves), physiological indicators reveal systemic stress and organ function (such as increased intracranial pressure indicating cerebral edema), behavioral data reflect motor function and nursing compliance (such as reduced activity may lead to pressure sores), and inflammatory data provide early warning of complications such as infection (such as PCT > 0.5 ng / mL indicating bacterial infection). The data acquisition module, through technological integration and precise processing, constructs a three-dimensional data network covering neurological function, systemic status, behavioral patterns, and inflammatory responses, providing underlying data support for solving the problems of fragmentation, lag, and simplification in current postoperative risk assessment.

[0030] The sub-risk assessment module is used to conduct sub-risk assessments on various indicator data and obtain various risk assessment results, including EEG assessment results, physiological indicator assessment results, behavioral assessment results, and inflammation assessment results. Each risk assessment result includes abnormal assessment results and normal assessment results.

[0031] The sub-risk assessment module includes an EEG signal risk assessment unit, a physiological indicator risk assessment unit, a behavioral risk assessment unit, and an inflammation risk assessment unit. Among them, the EEG signal risk assessment unit is used to identify abnormalities in EEG signal data based on an autoencoder and output EEG assessment results.

[0032] The process of an autoencoder identifying anomalies in EEG signal data is as follows: Obtaining periodically updated, anomaly-free EEG signal data from the database, and training the autoencoder based on this anomaly-free EEG signal data. The process is as follows:

[0033] The encoder in the autoencoder compresses the input abnormal EEG signal data into a low-dimensional EEG signal feature vector. The encoder consists of 2-3 fully connected layers and uses the ReLU activation function to compress high-dimensional EEG signals (such as a 1-second signal with a sampling rate of 1000Hz, corresponding to 1000-dimensional data) into low-dimensional feature vectors (such as 32-dimensional), achieving "feature reduction and noise filtering". The decoder in the autoencoder reconstructs the low-dimensional EEG signal feature vector back to the original dimension, obtaining the reconstructed abnormal EEG signal data. The decoder is symmetrical with the encoder and reconstructs the low-dimensional vector back to the original dimension signal through deconvolution or fully connected layers. The Sigmoid activation function is used to ensure that the output value is within a reasonable range.

[0034] The mean squared error is calculated based on the normal EEG signal data and the reconstructed normal EEG signal data. The loss function value is backpropagated to the parameters of the encoder and decoder through the backpropagation algorithm. The parameters are adjusted by the gradient descent method to continuously reduce the mean squared error. When the mean squared error is less than the set mean squared error threshold, the model is considered to have learned the feature distribution of normal EEG, the training ends, and the trained encoder is output, where the parameters include weights and biases.

[0035] The EEG signal data is input into the trained encoder, and after encoding and decoding, the reconstructed EEG signal data to be verified is obtained. The mean square error to be verified is calculated based on the reconstructed EEG signal data and the EEG signal data. If the mean square error to be verified is greater than the set error threshold, the output EEG evaluation result is abnormal; if the mean square error to be verified is not greater than the set error threshold, the output EEG evaluation result is normal.

[0036] The autoencoder automatically extracts core features of normal EEG (such as frequency band energy ratio and inter-electrode synchrony) through unsupervised learning and is sensitive to subtle changes that deviate from the normal pattern (such as a sudden 15% increase in delta wave energy and a significant increase in alpha wave phase difference). It can detect early abnormalities that are difficult to detect with traditional manual analysis (such as localized slow wave enhancement in EEG within 2 hours after surgery), providing early warning several hours before the onset of clinical symptoms, thus gaining golden time for intervention.

[0037] Training data comes from abnormal EEG data of patients undergoing similar surgeries. The model adaptively learns the normal baseline of specific populations (e.g., lower alpha wave frequency in elderly patients and more active beta waves in younger patients), avoiding misjudgments based on a "one-size-fits-all" standard. The EEG signal risk assessment unit, based on an autoencoder, solves the problems of traditional EEG monitoring—"high reliance on manual intervention, poor timeliness, and insufficient individual adaptability"—through data-driven intelligent analysis. It provides accurate and efficient technical support for monitoring neurological function in post-neurosurgery patients, significantly improving early risk identification capabilities and helping to achieve the clinical goal of "early detection, early intervention, and improved prognosis."

[0038] The physiological indicator risk assessment unit is used to obtain the physiological indicator assessment coefficient based on the physiological indicator data, and compare the physiological indicator assessment coefficient with the physiological indicator assessment threshold stored in the database: if the physiological indicator assessment coefficient is greater than the physiological indicator assessment threshold stored in the database, the physiological indicator assessment result is output as abnormal; if the physiological indicator assessment coefficient is not greater than the physiological indicator assessment threshold stored in the database, the physiological indicator assessment result is output as normal.

[0039] Basic vital signs parameters include heart rate ,blood pressure respiratory rate and body temperature Neurosurgical parameters include intracranial pressure. and blood oxygen saturation The process of obtaining physiological index evaluation coefficients based on physiological index data is as follows: Obtain the patient's basic pathological information, including age, gender, weight, height, and underlying disease information; obtain the basic case information of historical patients who have undergone the same neurosurgical operation stored in the case database (which stores the basic pathological information and postoperative physiological index data of patients who have undergone the same neurosurgical operation (such as brain tumor resection, cerebral hemorrhage surgery), forming a multidimensional feature space); analyze the similarity between the patient's basic pathological information and the basic case information of historical patients based on cosine similarity analysis, and determine the basic case information of historical patients that is most similar to the patient's basic pathological information.

[0040] Historical patient case information includes basic vital signs parameters and neurosurgical specific parameters. Basic vital signs parameters include heart rate. ,blood pressure respiratory rate and body temperature Historical neurosurgical parameters for patients include intracranial pressure. and blood oxygen saturation .

[0041] Based on the basic information of the most similar historical patient cases, the corresponding postoperative physiological index records stored in the database are obtained, including the postoperative physiological index data for each day after surgery. The physiological index data of the day is then compared with the corresponding postoperative physiological index data in the patient's postoperative physiological index records (the physiological index data of the current patient on the nth day after surgery (e.g., heart rate 85 beats / min, intracranial pressure 150 mmHg) is aligned with the data of the matching historical cases on the nth day after surgery (e.g., historical average heart rate 78 beats / min, intracranial pressure 130 mmHg) in terms of time dimension), and the physiological index evaluation coefficient is obtained.

[0042] Among them, the physiological index evaluation coefficient The calculation formula is as follows:

[0043] ;

[0044] and All are first relay functions. for The weighting factor of the function value, for The weighting factor of the function value. Taking the physiological indicator assessment scenario as an example, by calculating the relative difference between different physiological parameters and historical / standard values, the deviation of individual physiological data can be accurately captured; Using Euclidean distance, the differences in multidimensional data such as neurosurgical parameters are quantified; after weighted integration, the risk of physiological indicators can be measured in a personalized way, transforming vague "data anomalies" into calculable and comparable quantitative values, making the risk characteristics of complex physiological / signal data explicit, breaking through the traditional "one-size-fits-all" standard, fitting individual pathological characteristics, improving the accuracy of assessment, and reducing misjudgments and omissions.

[0045] The physiological indicator risk assessment unit, through personalized baseline construction and dynamic difference analysis, solves the shortcomings of traditional physiological monitoring such as "one-size-fits-all" and "static thresholds," enabling precise characterization of the physiological state of neurosurgical patients after surgery. It provides a scientific basis for the early detection of key risks such as circulatory disorders and abnormal intracranial pressure, helps medical staff develop individualized nursing plans, and improves the precision of postoperative monitoring.

[0046] The behavioral risk assessment unit is used to obtain behavioral assessment coefficients based on behavioral data and compare the behavioral assessment coefficients with the behavioral assessment thresholds stored in the database: if the behavioral assessment coefficient is greater than the behavioral assessment threshold stored in the database, the output behavioral assessment result is abnormal; if the behavioral assessment coefficient is not greater than the behavioral assessment threshold stored in the database, the output behavioral assessment result is normal.

[0047] The process of obtaining behavioral evaluation coefficients based on behavioral data is as follows: Obtain postoperative behavioral data of patients undergoing similar neurosurgical procedures; standardize the behavioral data and postoperative behavioral data and then perform cosine similarity analysis to obtain behavioral evaluation coefficients.

[0048] The process of obtaining assessment coefficients based on behavioral data starts with accurately identifying individual behavioral differences. In the risk assessment and intervention of neurosurgical postoperative care, it plays a role in assisting accurate judgment, supporting personalized care, and improving comprehensive assessment, which helps to improve the quality of postoperative care and the patient's recovery.

[0049] The inflammation risk assessment unit is used to obtain an inflammation assessment coefficient based on inflammation data and compare the inflammation assessment coefficient with the inflammation assessment threshold stored in the database: if the inflammation assessment coefficient is greater than the inflammation assessment threshold stored in the database, the output inflammation assessment result is abnormal; if the inflammation assessment coefficient is not greater than the inflammation assessment threshold stored in the database, the output inflammation assessment result is normal.

[0050] Blood inflammatory parameters include the levels of various white blood cells. C-reactive protein content and procalcitonin content 'i' represents the type of white blood cell; inflammatory mediator parameters include interleukin-6 levels. and tumor necrosis factor-α content ;like Figure 2 As shown, the process of obtaining the inflammation assessment coefficient based on inflammation data is as follows: Obtain normal human inflammation data, including normal human blood inflammation parameters and normal human inflammatory mediator parameters; normal human blood inflammation parameters include the normal content of various white blood cells. Normal C-reactive protein levels and normal levels of procalcitonin Normal inflammatory mediator parameters in the human body include the normal levels of interleukin-6. and normal levels of tumor necrosis factor-α .

[0051] Normal human inflammation data comes from a large-scale medical testing database of healthy individuals. Through long-term tracking and testing of people of different ages, genders, and physical conditions, statistical analysis is used to determine the normal reference ranges for various white blood cell types (such as neutrophils, lymphocytes, etc., corresponding to different I values), C-reactive protein, procalcitonin, interleukin-6, tumor necrosis factor-α, and other indicators, serving as a basic benchmark for judging whether inflammation is abnormal. For example, the normal neutrophil count in healthy adults is usually within a certain range. This data is based on statistical analysis of test results from a large number of physical examination populations and can reflect the homeostasis of inflammatory indicators in the human body when it is not subjected to pathological stimulation.

[0052] Postoperative inflammatory data were obtained from patients undergoing similar neurosurgical procedures, including postoperative blood inflammatory parameters and postoperative inflammatory mediator parameters. Postoperative blood inflammatory parameters included the levels of various white blood cells, C-reactive protein, and procalcitonin. Postoperative inflammatory mediator parameters included the levels of interleukin-6 and tumor necrosis factor-α.

[0053] Postoperative inflammatory data from patients undergoing similar neurosurgical procedures were selected from the hospital's medical record system. Historical cases matching the target patient's surgical type (e.g., brain tumor resection, cerebral hemorrhage evacuation, etc.), degree of surgical trauma (e.g., size of the bone window in craniotomy, influence of minimally invasive surgical approach), and patient baseline characteristics (age, presence of underlying diseases, etc.) were extracted. Blood inflammatory parameters and inflammatory mediator parameters from these cases were then extracted. These data reflect the changing patterns of postoperative inflammatory markers in patients undergoing similar procedures and can serve as a similar reference for assessing whether postoperative inflammation in the target patient is abnormal.

[0054] The first inflammation assessment factor is obtained based on normal human inflammation data and inflammation data; the second inflammation assessment factor is obtained based on postoperative inflammation data and inflammation data of patients undergoing similar neurosurgical procedures; the first and second inflammation assessment factors are weighted and summed to obtain the inflammation assessment coefficient.

[0055] Taking the first inflammation assessment factor as an example, its calculation formula is as follows:

[0056] ;

[0057] and All are second transfer functions. for The weighting factor of the function value, for The weighting factor of the function value.

[0058] The formula eliminates the influence of dimensions by using relative differences, and unifies the calculation logic of different types of data; at the same time, it covers a variety of indicators (from routine physiological to specific postoperative indicators), adapting to the complex clinical data environment.

[0059] The second inflammation assessment factor has the same form as the first inflammation assessment factor.

[0060] The comprehensive risk assessment module is used to merge the data of various indicators when all risk assessment results are normal, obtain a trained comprehensive risk assessment model, process the merged data of various indicators based on the trained comprehensive risk assessment model, and obtain the comprehensive risk assessment result, including the comprehensive assessment result being abnormal and the comprehensive assessment result being normal.

[0061] The comprehensive risk assessment model is a gradient boosting decision tree model, consisting of multiple decision trees (e.g., 100 or 200 trees). During training, each decision tree learns the mapping relationship between features and risk based on historical patients' comprehensive indicator vectors and corresponding risk labels (normal / abnormal). When a new comprehensive indicator vector is input, each decision tree judges the vector according to its internal splitting rules (e.g., selecting feature splitting points based on the Gini coefficient, mean squared error, etc.) and outputs a predicted value (this value can be understood as the probability contribution of the decision tree in considering the input vector to belong to the "normal" category). For example, a decision tree, based on the high proportion of delta wave energy in EEG features, judges a certain abnormal risk and outputs a predicted value of 0.3 (representing that the tree considers the probability of it belonging to the normal category to be 0.3).

[0062] The process of obtaining the comprehensive risk assessment result based on the comprehensive risk assessment model is as follows: The data of each indicator are processed to remove dimensions (normalization or standardization), and then concatenated into a comprehensive indicator vector. The concatenation logic is to sort the indicator data according to their category or importance, and then concatenate the dimensionless indicator data sequentially. For example, first concatenate the feature vector after EEG signal processing, then concatenate the physiological indicator vector, behavioral indicator vector, and inflammatory indicator vector to form a comprehensive indicator vector with unified dimensions.

[0063] The comprehensive index vector is input into the comprehensive risk assessment model to obtain the predicted value output by each decision tree. The predicted values ​​are accumulated according to the set weights to obtain the probability that the comprehensive index vector belongs to the normal range. If the probability is greater than the set normal probability threshold, the comprehensive risk assessment result is output as the comprehensive assessment result is normal. If the probability is not greater than the set normal probability threshold, the comprehensive risk assessment result is output as the comprehensive assessment result is abnormal.

[0064] Postoperative risks in neurosurgery are influenced by a synergistic effect of multiple factors, including electroencephalography (EEG), physiological function, behavior, and inflammation. A normal single indicator does not guarantee overall risk control. The comprehensive risk assessment module integrates multi-dimensional indicators and utilizes a GBDT model to uncover complex correlations between indicators (such as the synergistic risk between abnormal EEG and elevated inflammatory markers). For example, even if EEG and physiological indicators are normal, but behavioral indicators show a persistent decline in activity tolerance and a potential upward trend in inflammatory markers, the comprehensive model can identify this "latent" risk, preventing the overlooking of overall risk due to a single normal indicator and improving the comprehensiveness of risk assessment. By applying multi-dimensional data fusion and a gradient-enhanced decision tree model, the module addresses the limitations of single-indicator assessments in neurosurgery, achieving precise quantitative assessment of comprehensive risk. This provides strong support for medical staff to develop personalized and precise postoperative care and intervention plans, helping to improve patient recovery quality and risk management effectiveness.

[0065] The early warning module is used to obtain the results of various risk assessments and the comprehensive risk assessment, and to issue early warnings based on these results.

[0066] The process of issuing early warnings based on the results of various risk assessments and the comprehensive risk assessment is as follows: Count the number of abnormal assessment results obtained from the sub-risk assessments. If the number is no more than two, a Level 1 warning is issued; if the number is more than two, a Level 2 warning is issued. If an abnormal comprehensive assessment result is received, a Level 2 warning is issued directly.

[0067] A Level 1 warning indicates that the patient has localized, relatively independent risk points after surgery, which have not yet formed a multi-dimensional collaborative risk. For example, only abnormal EEG signals (which may be transient brain function fluctuations) or only abnormal physiological indicators (such as a single heart rate fluctuation) are present. Although these situations require attention, their risk diffusion and urgency are relatively low. When the system triggers a Level 1 warning, it can remind medical staff to conduct targeted examinations through methods such as flashing yellow warning lights in the ward and pop-up notifications on the nurses' station computer (marking the specific abnormal items).

[0068] The Level 2 warning includes two triggering scenarios. First, if the number of abnormalities in the sub-risk assessments is greater than 2, it indicates that the patient faces risks across multiple dimensions post-surgery, including EEG, physiology, behavior, and inflammation, with strong risk synergy, potentially leading to serious post-operative complications (such as multiple organ dysfunction syndrome). Second, if the overall risk assessment result is abnormal, even if the number of sub-item abnormalities is small, the model judges an overall high risk through multi-dimensional data fusion (e.g., individual indicators may not show significant abnormalities, but their synergistic effect suggests potential serious problems). When a Level 2 warning is triggered, strong reminders are used, such as a constantly lit red warning light in the ward, audible and visual alarms at the nurses' station, and emergency messages pushed to the on-duty mobile phone, to urge medical staff to respond immediately and conduct multidisciplinary collaborative assessments and interventions.

[0069] An electronic device includes: a processor; and a memory storing computer program instructions that, when executed by the processor, cause the processor to perform the postoperative care risk warning system for neurosurgical patients as described above.

[0070] A computer-readable storage medium for storing a program that, when executed by a processor, implements the postoperative care risk warning system for neurosurgical patients as described above.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0075] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0076] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A postoperative nursing risk early warning system for neurosurgical patients, characterized in that, It includes a data acquisition module, a component risk assessment module, a comprehensive risk assessment module, and an early warning module, among which: The data acquisition module is used to acquire various postoperative indicator data of neurosurgical patients, including electroencephalogram (EEG) signal data, physiological indicator data, behavioral data, and inflammation data. The sub-risk assessment module is used to conduct sub-risk assessments on various indicator data and obtain various risk assessment results, including EEG assessment results, physiological indicator assessment results, behavioral assessment results, and inflammation assessment results. Each risk assessment result includes abnormal assessment results and normal assessment results. The comprehensive risk assessment module is used to merge the data of various indicators when all risk assessment results are normal, obtain a trained comprehensive risk assessment model, process the merged data of various indicators based on the trained comprehensive risk assessment model, and obtain the comprehensive risk assessment result, including the comprehensive assessment result being abnormal and the comprehensive assessment result being normal. The early warning module is used to obtain the results of various risk assessments and the comprehensive risk assessment, and to issue early warnings based on these results. The sub-risk assessment module includes a brainwave signal risk assessment unit, a physiological indicator risk assessment unit, a behavioral risk assessment unit, and an inflammation risk assessment unit; The process of issuing early warnings based on the results of various risk assessments and the comprehensive risk assessment is as follows: If the number of abnormal assessment results obtained from the statistical sub-risk assessment is no more than two, a Level 1 warning will be issued; if the number is more than two, a Level 2 warning will be issued. If the received comprehensive assessment results are abnormal, a Level 2 warning will be issued directly.

2. The postoperative nursing risk early warning system for neurosurgical patients according to claim 1, characterized in that, The data acquisition module includes an electroencephalogram (EEG) signal acquisition unit, a physiological indicator acquisition unit, a behavioral data acquisition unit, and an inflammation data acquisition unit, among which: The EEG signal acquisition unit is used to connect to a high-density EEG device to acquire initial EEG signals from multiple electrode sites throughout the brain, and to process the initial EEG signals based on independent component analysis and bandpass filtering to obtain EEG signal data. The physiological index acquisition unit is used to acquire the basic physiological indexes of neurosurgical patients after surgery. The basic physiological indexes are preprocessed and then standardized to obtain physiological index data. The behavior data acquisition unit is used to collect patient action data based on the camera, identify the patient action data based on the target detection algorithm, determine the patient behavior, and acquire behavior data based on the inertial measurement sensor. The inflammation data acquisition unit is used to acquire the patient's postoperative clinical examination results, extract the initial inflammation data, and obtain the inflammation data after standardizing the initial inflammation data.

3. The postoperative nursing risk early warning system for neurosurgical patients according to claim 2, characterized in that: EEG signal data includes EEG frequency band parameters, EEG time domain parameters, and EEG spatial parameters; Physiological data include basic vital signs parameters and neurosurgical specific parameters; Behavioral data includes the acceleration and angular velocity during movement corresponding to each patient's behavior; Inflammation data includes blood inflammatory parameters and inflammatory mediator parameters.

4. The postoperative nursing risk early warning system for neurosurgical patients according to claim 3, characterized in that: The EEG signal risk assessment unit is used to identify anomalies in EEG signal data based on an autoencoder and output EEG assessment results. The physiological indicator risk assessment unit is used to obtain physiological indicator assessment coefficients based on physiological indicator data, and then compares these coefficients with the physiological indicator assessment thresholds stored in the database. If the physiological indicator evaluation coefficient is greater than the physiological indicator evaluation threshold stored in the database, the physiological indicator evaluation result will be output as abnormal. If the physiological indicator evaluation coefficient is not greater than the physiological indicator evaluation threshold stored in the database, the output physiological indicator evaluation result is normal. The behavioral risk assessment unit is used to obtain behavioral assessment coefficients based on behavioral data, and then compares these coefficients with behavioral assessment thresholds stored in the database. If the behavior evaluation coefficient is greater than the behavior evaluation threshold stored in the database, the output behavior evaluation result will be "abnormal evaluation result". If the behavior evaluation coefficient is not greater than the behavior evaluation threshold stored in the database, the output behavior evaluation result is "normal". The inflammation risk assessment unit is used to obtain an inflammation assessment coefficient based on inflammation data, and then compares the inflammation assessment coefficient with the inflammation assessment thresholds stored in the database. If the inflammation assessment coefficient is greater than the inflammation assessment threshold stored in the database, the output inflammation assessment result will be "abnormal assessment result". If the inflammation assessment coefficient is not greater than the inflammation assessment threshold stored in the database, the output inflammation assessment result is "normal".

5. The postoperative nursing risk early warning system for neurosurgical patients according to claim 4, characterized in that, The process by which an autoencoder identifies anomalies in EEG signal data is as follows: Obtain abnormal EEG signal data from the database that is updated regularly. The autoencoder is trained based on abnormal EEG signal data, and the process is as follows: The encoder in the autoencoder compresses the input abnormal EEG signal data into a low-dimensional EEG signal feature vector, and the decoder in the autoencoder reconstructs the low-dimensional EEG signal feature vector back to the original dimension to obtain the reconstructed abnormal EEG signal data. The mean squared error is calculated based on the abnormal EEG signal data and the reconstructed abnormal EEG signal data. The value of the loss function is backpropagated to the parameters of the encoder and decoder through the backpropagation algorithm. The parameters are adjusted by the gradient descent method so that the mean squared error is continuously reduced. When the mean squared error is less than the set mean squared error threshold, the training ends and the trained encoder is output, where the parameters include weights and biases. The EEG signal data is input into the trained encoder, and after encoding and decoding, the reconstructed EEG signal data to be verified is obtained. The mean square error to be verified is calculated based on the reconstructed EEG signal data and the EEG signal data. If the mean square error to be verified is greater than the set error threshold, the output EEG assessment result will be considered abnormal. If the mean square error to be verified is not greater than the set error threshold, the output EEG assessment result is normal.

6. The postoperative nursing risk early warning system for neurosurgical patients according to claim 4, characterized in that, Basic vital signs parameters include heart rate, blood pressure, respiratory rate, and body temperature; neurosurgical specific parameters include intracranial pressure and blood oxygen saturation. The process of obtaining physiological indicator evaluation coefficients based on physiological indicator data is as follows: Obtain the patient's basic pathological information, including age, gender, weight, height, and information on underlying diseases; Obtain basic information of historical patients who have undergone the same neurosurgical procedure from the case database. Analyze the similarity between the patient's pathological information and the historical patient's basic information based on cosine similarity to determine the historical patient's basic information that is most similar to the patient's pathological information. Based on the most similar historical patient case information, retrieve the corresponding postoperative physiological index records stored in the database, including daily postoperative physiological index data. The physiological index data of the day is compared with the corresponding postoperative physiological index data in the patient's postoperative physiological index record data to obtain the physiological index evaluation coefficient.

7. The postoperative nursing risk early warning system for neurosurgical patients according to claim 4, characterized in that, The process of obtaining behavioral evaluation coefficients based on behavioral data is as follows: Obtain postoperative behavioral data from patients undergoing similar neurosurgical procedures; After standardizing the behavioral data and postoperative behavioral data, cosine similarity analysis was performed to obtain the behavioral evaluation coefficient.

8. The postoperative nursing risk early warning system for neurosurgical patients according to claim 4, characterized in that, Blood inflammatory parameters include the levels of various white blood cells, C-reactive protein, and procalcitonin; inflammatory mediator parameters include the levels of interleukin-6 and tumor necrosis factor-α. The process of obtaining the inflammation assessment coefficient based on inflammation data is as follows: Acquire normal human inflammatory data, including normal blood inflammatory parameters and normal inflammatory mediator parameters; Obtain postoperative inflammatory data from patients undergoing similar neurosurgical procedures, including postoperative blood inflammatory parameters and postoperative inflammatory mediator parameters; The first inflammation assessment factor is obtained based on normal human inflammation data and inflammation data. A second inflammation assessment factor was obtained based on postoperative inflammation data and inflammation data of patients undergoing similar neurosurgical procedures. The inflammation assessment coefficient is obtained by weighted summation of the first and second inflammation assessment factors.

9. The postoperative nursing risk early warning system for neurosurgical patients according to claim 2, characterized in that, The comprehensive risk assessment model is a gradient boosting decision tree model. The process of obtaining the comprehensive risk assessment result based on the comprehensive risk assessment model is as follows: The data of each indicator are processed to remove dimensions and then concatenated into a comprehensive indicator vector. The comprehensive index vector is input into the comprehensive risk assessment model to obtain the predicted value output by each decision tree. The predicted values ​​are accumulated according to the set weights to obtain the probability that the comprehensive index vector belongs to the normal range. If the probability is greater than the set normal probability threshold, the output of the comprehensive risk assessment result is "comprehensive assessment result is normal". If the probability is not greater than the set normal probability threshold, the output comprehensive risk assessment result will be "comprehensive assessment result abnormal".

Images