Intraoperative hypotension early warning method and related equipment

Abstract

The invention provides an intraoperative hypotension early warning method and related equipment, and is applied to the technical field of data processing. The method comprises the following steps: processing a preset hypotension early warning model based on a training sample set with identification information to generate a target hypotension early warning model; processing the operation information of the target user in a preset time period, and generating an interval prediction sub-task and an intra-operation prediction sub-task; processing the interval prediction subtask and the intra-operative prediction subtask to generate a preset prediction feature vector; processing the physiological motion parameter information of the target user to generate a blood pressure limiting factor of the target user; processing the physiological state information of the target user to generate an ambulatory blood pressure influence factor of the target user; and processing the preset prediction feature vector, the blood pressure limiting factor of the target user and the ambulatory blood pressure influence factor of the target user based on the target hypotension early warning model to generate hypotension early warning information of the target user.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an early warning method for intraoperative hypotension and related equipment. Background Art

[0002] Intraoperative hypotension is associated with adverse patient outcomes, including postoperative acute kidney injury, myocardial infarction, long-term morbidity, and death. Risk factors include male sex, advanced age, American Society of Anesthesiologists (ASA) class IV, emergency surgery, and pre-induction hypotension. However, there is currently no universal standard for defining intraoperative hypotension.

[0003] Prolonged hypotension is associated with adverse postoperative outcomes. Previous studies have shown that short-term hypotensive events are still associated with an increased risk of postoperative complications. Although risk factors for inducing intraoperative hypotension are known, they do not reduce its incidence or duration. Recently, real-time prediction models for predicting hypotension have been developed by analyzing patient monitoring parameters. However, existing clinical studies have yielded conflicting conclusions on whether the application of these prediction models can optimize patient outcomes.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the Invention

[0005] The purpose of this application is to provide an early warning method and related equipment for intraoperative hypotension, which at least to some extent overcomes the problems existing in the prior art. By obtaining surgery, physiological parameters, status information, sample sets and models, and constructing a sample set optimization model with hypotension identification after sample preprocessing. The surgical information is processed to obtain prediction subtasks and feature vectors, and the physiological parameters are analyzed to obtain blood pressure limits and influencing factors. Finally, the multiple factors are integrated through matrix processing and calibration to output hypotension early warning details, covering the risk level, time period, inducement and intervention measures, forming a complete early warning closed loop, providing solid scientific support for the prevention and control of intraoperative hypotension, effectively ensuring surgical safety and improving user prognosis.

[0006] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0007] According to one aspect of the present application, a warning method for intraoperative hypotension is provided, including: obtaining surgical information, physiological motion parameter information, and physiological status information of a target user within a preset time period; processing the surgical information of the target user within the preset time period to generate an interval prediction subtask and an intraoperative prediction subtask; processing the interval prediction subtask and the intraoperative prediction subtask to generate a preset prediction feature vector; processing the physiological motion parameter information of the target user to generate a blood pressure limitation factor of the target user; processing the physiological status information of the target user to generate a dynamic blood pressure influencing factor of the target user; processing the preset prediction feature vector, the blood pressure limitation factor of the target user, and the dynamic blood pressure influencing factor of the target user based on a target hypotension warning model to generate hypotension warning information for the target user.

[0008] Another aspect of the present application is a warning device for intraoperative hypotension, comprising: an acquisition module for acquiring surgical information of a target user within a preset time period, physiological motion parameter information of the target user, and physiological status information of the target user; a processing module for processing the surgical information of the target user within the preset time period to generate an interval prediction subtask and an intraoperative prediction subtask; processing the interval prediction subtask and the intraoperative prediction subtask to generate a preset prediction feature vector; processing the physiological motion parameter information of the target user to generate a blood pressure limitation factor of the target user; processing the physiological status information of the target user to generate a dynamic blood pressure influencing factor of the target user; processing the preset prediction feature vector, the blood pressure limitation factor of the target user, and the dynamic blood pressure influencing factor of the target user based on a target hypotension warning model to generate hypotension warning information for the target user.

[0009] According to another aspect of the present application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to implement the above-mentioned early warning method for intraoperative hypotension by executing the executable instructions.

[0010] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a second processor, the above-mentioned early warning method for intraoperative hypotension is implemented.

[0011] This application provides a method and related equipment for early warning of intraoperative hypotension. The server obtains surgical procedures, physiological parameters, status information, sample sets, and models, and constructs a sample set optimization model with hypotension identification through sample preprocessing. The surgical information is processed to obtain prediction subtasks and feature vectors, and physiological parameters are analyzed to obtain blood pressure limits and influencing factors. Finally, multiple factors are integrated through matrix processing and calibration to output hypotension early warning details, covering risk level, time period, inducements, and intervention measures, forming a complete early warning closed loop, providing solid scientific support for the prevention and control of intraoperative hypotension, effectively ensuring surgical safety and improving user prognosis.

[0012] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart showing a method for early warning of intraoperative hypotension provided by an embodiment of the present application is shown;

[0014] Figure 2 A schematic structural diagram of an early warning device for intraoperative hypotension provided in one embodiment of the present application is shown. DETAILED DESCRIPTION

[0015] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0016] The following combination Figure 1 To describe the early warning method for intraoperative hypotension according to an exemplary embodiment of the present application. It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited in this respect. On the contrary, the embodiments of the present application can be applied to any applicable scenario.

[0017] In one embodiment, Figure 1 The following schematically illustrates a flow chart of a method for early warning of intraoperative hypotension according to an embodiment of the present application. The method is applied to a server and includes:

[0018] S101, obtaining surgical information of a target user within a preset time period, physiological motion parameter information of the target user, and physiological status information of the target user.

[0019] In one embodiment, the surgical schedule for the target user in the preset time period (1 week before surgery to 72 hours after surgery) is accurately extracted from the hospital's surgical management system. For example, information on radical resection of gastrointestinal tumors in general surgery is obtained, the surgical level is level III, and the planned duration is 4 hours. Looking back on the patient's surgical history in the past 5 years, including 2 appendectomies, he had a complication of transient intestinal obstruction after surgery, and the recovery period was 2 weeks. The intraoperative operation interference with circulation is estimated based on the type and complexity of the surgery. Historical surgical complications help analyze potential hypotension risk tendencies, set a baseline for risk prediction, and help medical staff plan blood pressure management strategies and prepare emergency resources based on the characteristics of the surgery.

[0020] For example, for gastrointestinal surgery, intravenous-inhalation combined general anesthesia is used, with propofol, fentanyl, and cisatracurium used for induction, and sevoflurane, remifentanil, and intermittent muscle relaxants used for maintenance. The dosage is finely adjusted according to the user's age, physical condition, and the progress of the operation. The depth of anesthesia is monitored and maintained at 40-60 using BIS. Fluid management is based on the length of fasting and the estimated amount of blood loss during the operation. The first 1000ml of crystalloid Ringer's solution is supplemented, and colloid fluids and blood products are prepared. Colloids or blood transfusions are given in a timely manner according to the amount of blood loss. The cardiovascular effects of anesthetic drugs and fluid balance strategies are closely related to hypotension. Precision anesthesia planning helps analyze the impact of anesthetic factors on the blood pressure curve, construct an anesthesia-hypotension risk model, improve the predictability of blood pressure fluctuations, and ensure stable blood pressure during anesthesia.

[0021] Physiological parameters are as follows: The bedside monitor collects data at high frequency. Cardiac monitoring shows a heart rate of 70-90 beats / min, with occasional premature atrial contractions of 2 / hour, and a stable waveform. Respiratory rate is 12-18 breaths / min with a regular rhythm, and oxygen saturation fluctuates between 95% and 98%. Arterial blood pressure waveform analysis reveals systolic pressure of 110-130 mmHg, diastolic pressure of 70-80 mmHg, mean arterial pressure of 85-95 mmHg, and pulse pressure of 30-50 mmHg, with steady rates of pressure rise and fall. PET / CO2 averages 35-40 mmHg, with a stable slope during the expiratory rise and a steep slope during the inspiratory fall. Dynamic monitoring of cardiopulmonary parameters captures transient hypotension events caused by cardiopulmonary imbalance, such as abnormal breathing causing intrathoracic pressure changes that affect venous return, and sudden changes in heart rate that affect cardiac output and blood pressure. This provides precise parameters for real-time blood pressure control, helping medical staff to detect prelude signals of hypotension and intervene promptly.

[0022] The PPG sensor measures an average PTT of 120-140ms, an AI of 10%-15%, and a PI of 4-6. NIRS indicates dynamic stability of muscle HbO2 at 70%-80% and Hb at 20%-30%. PTT reflects vascular elasticity and peripheral resistance, AI indicates arterial stiffness, and PI and Hb dynamically reflect peripheral perfusion. For example, in elderly patients undergoing surgery, a prolonged PTT, elevated AI, and decreased PI before surgery indicate peripheral vascular degeneration, which can lead to intraoperative hypotension. Accurate assessment of peripheral circulatory parameters allows for pre-adjustment of fluids and vasoactive medications, optimizing volume management and maintaining vascular tone, enhancing the accuracy of hypotension prevention, maintaining stable peripheral perfusion and blood pressure, and improving surgical safety margins.

[0023] Physiological status information is as follows: 65-year-old male, BMI 25kg / m 2 He had hypertension for 10 years (taking nifedipine controlled-release tablets, blood pressure 130-140 / 80-90 mmHg), and diabetes for 5 years (using metformin to control blood sugar, HbA1c 7%). His blood test results were Hb 130 g / L, Hct 40%, biochemical glucose 7 mmol / L, serum creatinine 100 μmol / L, serum potassium 4 mmol / L, and normal liver function. His coagulation PT 12s, APTT 30s, and PLT 200×10 9 / L, blood gas pH7.4, PaO290mmHg, PaCO238mmHg. Synergistic analysis of chronic diseases and physiological indicators shows that hypertension causes vascular remodeling, diabetes damages microvessels and neural regulation, all of which increase the risk of intraoperative hypotension. According to individual physiological baseline, medical care can optimize perioperative management, adjust medication, correct metabolic disorders, stabilize the internal environment, reduce the incidence of hypotension, and improve the quality of surgical prognosis. Cardiac ultrasound: LVEF55%, CO4-5L / min, E / A1.2, EF slope is normal, indicating that the heart pumps blood well; pulmonary function FVC80%, FEV175%, FEV1 / FVC70%, DLCO85% predicted values, mild ventilation function reduction; renal function eGFR80ml / min / 1.73m 2 , negative urine protein; Child-Pugh class A liver disease, 15-minute ICG retention rate of 5%, good liver reserve. Multi-organ function assessment stratifies hypotension risk. Reduced lung function leads to decreased oxygenation and increased cardiac load, which can predispose to hypotension. Renal function affects fluid and electrolyte balance and drug excretion, indirectly disrupting blood pressure. Organ function quantification identifies physiological vulnerabilities, helping medical staff customize multidisciplinary collaborative management plans to prevent organ function deterioration from leading to hypotension, maintain circulatory stability and organ perfusion during surgery, improve surgical safety and success rates, and optimize the patient's recovery trajectory.

[0024] We collected massive surgical user data from multi-center medical databases, including records of intraoperative physiological parameters and hypotension incidence for users of different surgery types, age groups (18-85 years old), genders, and those with multiple underlying medical conditions (such as hypertension, coronary heart disease, diabetes, chronic obstructive pulmonary disease, and chronic kidney disease) over the past 10 years. This ensured sample diversity and representativeness, encompassing both common and complex surgical scenarios and populations with diverse physiological characteristics. This provided a rich data foundation for model learning, enhanced generalization capabilities, and accurately captured characteristic patterns of hypotension risk under various combinations of factors. We also accurately labeled the sample data. Based on the intraoperative arterial blood pressure monitoring values and the diagnostic criteria for hypotension (such as a drop in mean arterial pressure of more than 20% compared to the baseline value and lasting for more than 5 minutes), the occurrence of hypotension and key details such as the starting time, duration, and lowest blood pressure value are clearly marked; at the same time, key node information of the surgical process (operative stage, operation type, changes in anesthesia depth), the user's real-time physiological status (heart rate, heart rhythm, respiratory rate, blood oxygen saturation, central venous pressure, cardiac output, and use of vasoactive drugs) and laboratory test indicators (key values such as preoperative hemoglobin, blood creatinine, blood sugar, coagulation function, and dynamic changes after surgery) are entered to construct a high-dimensional, dynamic, labeled sample data set, providing a comprehensive and accurate learning example for the model, driving precise risk pattern mining and prediction rule construction.

[0025] Missing physiological parameters are filled in based on clinical experience and data distribution characteristics (e.g., mean, median, or predicted values based on regression models). Abnormal blood pressure values are corrected using professional knowledge and statistical methods (combined with heart rate, vasoactive drug dosage, and surgical procedure rationality judgment) to ensure data quality and reliability, prevent "dirty data" from interfering with model learning, and improve model stability and prediction accuracy. Physiological characteristics with different dimensions and value ranges in the sample set are normalized. Numerical features (e.g., blood pressure, heart rate, and laboratory index values) are normalized using the Z-score or Min-Max normalization methods to map feature values to specific intervals (e.g., [0, 1] or [-1, 1]), unify the dimensional scale, balance the contribution weight of each feature in model training, prevent model training bias due to differences in feature magnitude, and optimize model convergence efficiency and performance. Sample subsets are constructed by stratification and grouping based on surgical type, user age, and underlying disease category. For example, cardiac surgeries are subdivided by surgical method (valve replacement, coronary artery bypass grafting, heart transplantation); age groups are divided into 10-year intervals; and underlying diseases are categorized by disease type and severity (mild, moderate, and severe hypertension are graded based on blood pressure control levels and comorbidities). Data augmentation techniques (SMOTE algorithm oversampling minority samples, random data transformation, and noise augmentation) are used within each subset to improve sample balance and diversity, strengthen the model's learning of hypotension risk patterns in specific surgical populations and disease groups, avoid model bias due to sample imbalance, and enhance the model's predictive accuracy and robustness in different clinical scenarios.

[0026] The pre-set hypotension warning model comprehensively considers the complexity, nonlinearity, and individual differences of intraoperative hypotension risk factors, selecting a hybrid architecture model that integrates a deep neural network (DNN) and a Bayesian network (BN). DNN's powerful automatic feature learning and deep abstraction capabilities can mine hidden hypotension risk characteristic patterns and complex nonlinear relationships from massive amounts of high-dimensional surgical physiological data. For example, it can deeply analyze the deep mapping of surgical stimulation intensity, the cumulative effects of anesthetic drugs, and the dynamic coupling changes of physiological indicators to the occurrence of hypotension. BN, based on the advantages of probabilistic graphical models, effectively integrates clinical prior knowledge (such as the causal relationship between different diseases and surgical factors and hypotension determined by expert experience) with data-driven learning results to achieve precise quantification of hypotension risk probability and uncertainty reasoning, providing risk confidence intervals and explanatory basis for clinical decision-making. The integration of the two achieves a balance between accurate prediction and clinical interpretability, improving the model's clinical acceptance and application value. Model hyperparameters are reasonably set based on the selected architecture and sample data characteristics. Determine the number of DNN network layers (3-5 hidden layers are set according to data complexity), the number of neurons in each layer (the number of neurons in the first hidden layer is determined by the input feature dimension, and the number of neurons in subsequent layers is determined by the balance between learning ability and overfitting risk), activation function (such as the ReLU function to improve training efficiency and gradient propagation stability), learning rate (adaptive learning rate algorithm is used to initially set it to 0.001 and dynamically adjusted according to the training process), and regularization parameter (L1 / L2 regularization controls model complexity to prevent overfitting); the BN network constructs the network structure based on the causal relationship between surgical factors and physiological indicators, determines the initial parameters of the node conditional probability distribution, and lays the foundation for the model training optimization infrastructure.

[0027] A strategy combining unsupervised pre-training and supervised fine-tuning is used to optimize model performance. First, unsupervised pre-training of the DNN is performed using autoencoders (AE) or restricted Boltzmann machines (RBM) on large-scale unlabeled surgical physiological data. This allows for learning the data's underlying feature representation and distribution patterns, initializing network weights and improving the model's feature extraction and generalization capabilities. The model parameters are then fine-tuned using a backpropagation algorithm on a labeled sample set based on the supervised learning objective (minimizing the error in hypotension risk prediction). Clinical label information is then integrated to accurately calibrate the model's predictive capabilities. Multiple rounds of iterative optimization ensure that the model's performance on the validation set converges and stabilizes (e.g., accuracy exceeding 85%, recall exceeding 80%, and F1 value exceeding 0.8), ensuring that the model accurately captures hypotension risk characteristics and provides a reliable technical core for real-time intraoperative early warning.

[0028] In another embodiment, age groups are divided at intervals of 10 years, such as 18-27 years old, 28-37 years old, etc. For the elderly group (over 65 years old), their physiological characteristics are often accompanied by organ function decline, reduced cardiac reserve function leads to reduced heart rate variability, vascular sclerosis increases peripheral vascular resistance, and they are more susceptible to anesthesia and surgical operations during surgery to cause hypotension. Although the young group (18-44 years old) has stronger physiological functions, special diseases (such as Marfan syndrome causing cardiovascular disease) or long-term unhealthy lifestyles (heavy smoking causing endothelial damage) can also change physiological characteristics and increase the risk of hypotension. The physiological characteristic information such as the mean heart rate, blood pressure fluctuation range, respiratory function indicators (such as forced vital capacity, rate in one second) and metabolic and nutritional indicators (such as albumin level, blood lipid status) reflected by laboratory tests of users of all age groups during surgery are extracted from the database to form an age-stratified sample subset, which provides an accurate data cornerstone for the subsequent analysis of the impact of age factors on hypotension.

[0029] Classification by common chronic diseases, such as hypertension, coronary heart disease, diabetes, chronic obstructive pulmonary disease (COPD), and chronic kidney disease. Hypertensive users have long-term blood pressure fluctuations that lead to changes in the structure and function of the blood vessel walls, making blood pressure control during surgery difficult. Their sample subset contains information such as blood pressure circadian rhythm characteristics, types and dosages of antihypertensive drugs, and the degree of fundus vascular lesions to help analyze the risk of blood pressure instability; users with coronary heart disease have a high risk of imbalance between cardiac blood supply and oxygen supply and demand. The samples cover characteristics such as the degree of coronary artery stenosis, history of myocardial infarction, cardiac function classification, and myocardial enzyme indicators to help evaluate the association between myocardial oxygen consumption and hypotension during surgery; diabetic users have microvascular and neuropathy that affect vasoconstriction and blood volume regulation. The samples contain blood sugar fluctuation characteristics (such as glycated hemoglobin, blood sugar standard deviation), peripheral nerve conduction velocity, and foot skin microcirculation perfusion index. The sample collection of COPD users includes lung function indicators (such as residual volume, pulmonary diffusion function), blood gas analysis parameters (long-term trend of oxygen partial pressure and carbon dioxide partial pressure) and long-term oxygen therapy, and explores the causal chain of respiratory dysfunction leading to hypotension; chronic kidney disease users are stratified according to glomerular filtration rate, and the samples include dynamic renal function indicators, electrolyte disorder characteristics (such as abnormal frequency and amplitude of blood potassium and blood calcium), renal anemia indicators and dialysis treatment details, explaining the interactive influence path between renal function damage and hypotension. The samples of each complication group provide a solid data support framework for the targeted exploration of hypotension risk factors. Feature extraction is performed on the grouped training sample set to generate the original feature library. From the grouped sample set, the mean, median, standard deviation, extreme value and coefficient of variation of physiological indicators are calculated. For example, in the cardiac surgery user group, the mean heart rate reflects the basal metabolism and cardiac stress level, and the standard deviation of the heart rate measures the stability of the heart rhythm. During surgery, an excessively high mean heart rate or an increased standard deviation indicates excessive sympathetic nerve excitation, which can easily lead to hypotension. Extreme blood pressure values help determine the individual blood pressure fluctuation range. A large coefficient of variation in systolic blood pressure means high sensitivity to blood pressure regulation and an increased risk of hypotension under surgical stress. The mean and standard deviation of respiratory rate and depth are related to pulmonary ventilation efficiency and oxygen supply stability. Abnormal fluctuations indicate that respiratory dysfunction affects intrathoracic pressure and venous return, causing hypotension, thereby uncovering potential hypotension risk signals.

[0030] In the process of developing a hypotension early warning system through joint analysis of multiple waveforms, it is necessary to comprehensively exploit the key physiological and pathological information contained in each waveform. For the ECG waveform, the P wave duration and morphology parameters must be precisely extracted. P wave dispersion, acting as a precise ruler, measures the consistency of atrial conduction, providing key clues for assessing the stability of the initial phase of cardiac electrical activity. Stable and regular atrial conduction is the cornerstone of maintaining normal blood pressure. Abnormal conduction can have ripple effects throughout the body, quietly laying the groundwork for hypotension. The duration and morphology of the QRS complex must be closely monitored. Any widening or notching of the QRS complex, like a malfunctioning traffic light, signals a sudden increase in the risk of myocardial block. This directly disrupts the heart's core pumping mechanism, leading to unstable stroke volume (SV) output and a sharp decrease in effective cardiac ejection, potentially triggering a hypotensive crisis. High-resolution analysis of ST segment morphology, including the magnitude of depression or elevation, as well as the precise measurement of its duration, acts as a pinpointer, pinpointing the time, scope, and severity of myocardial ischemia. Myocardial ischemia is like a shortage of raw materials in a factory, limiting the heart's powerhouse and reducing stroke volume. This is closely linked to hypotension, acting as a causal force. Constantly monitor changes in T-wave morphology and amplitude. T-waves that are high, inverted, or flat are like a barometer on an electrocardiogram (ECG), reflecting abnormal myocardial repolarization. These abnormalities, like turbulent currents in an electrical circuit, disrupt cardiac electrophysiological stability, impairing cardiac rhythm and contraction coordination, leading to fluctuations in stroke volume and creating the conditions for hypotension.

[0031] The arterial pressure waveform is meticulously analyzed for systolic, diastolic, and mean arterial pressure. Its slope rate of change acts like an accelerometer for blood pressure dynamics. A steep decline indicates a serious imbalance in the coordination between vascular elasticity and cardiac contraction. Poor vascular elasticity prevents adequate support for cardiac contraction, forcing the heart to pump blood against greater resistance. This makes it difficult to maintain a stable stroke volume, leading to a sharp increase in the risk of hypotension. As a barometer of the dynamic balance between blood volume and vascular tone, any abnormal fluctuation in pulse pressure variation, like ripples on the surface of a lake, signals a breakdown in blood volume management or vascular tone regulation, leading to uncertainty in cardiac output and a consequent increase in the risk of hypotension. Detailed analysis of changes in the depth of the dicrotic notch reveals that it acts like a display of the status of the peripheral vascular "regulatory valve," reflecting the interplay between peripheral vascular resistance regulation and arterial compliance. Abnormal fluctuations in peripheral vascular resistance are directly linked to systemic vascular resistance (SVR), affecting cardiac afterload, disrupting stroke volume output, and undermining blood pressure stability, potentially contributing to hypotension. Real-time tracking of stroke volume (SV) and stroke volume variation (SVV) is also available. The stroke volume directly reflects the amount of blood ejected by each heartbeat. A stable and sufficient stroke volume is the key to maintaining blood pressure. The stroke volume variability is like a "dynamic monitor", sensitively capturing the successive fluctuations in the stroke volume. Once SVV exceeds the normal threshold, it indicates that the heart's pumping function is unstable, which may be due to factors such as insufficient blood volume, decreased heart function or vascular abnormalities, leading to the risk of hypotension, providing a key warning for timely intervention.

[0032] For photoplethysmography, pulse wave transit time and its coefficient of variation are precisely extracted, acting like a vascular "elastic pulse," assessing the dynamic changes in vascular elasticity and peripheral resistance. Prolonged transit time or significant fluctuations indicate decreased vascular elasticity and an imbalance in peripheral resistance. This increases the resistance the heart must overcome to pump blood, resulting in decreased stroke volume and the imminent risk of hypotension. Closely monitoring the perfusion index and its coefficient of variation serves as a "barometer" of blood supply to peripheral tissues, reflecting the level of perfusion. Low values or frequent fluctuations indicate unstable peripheral blood supply, potentially due to insufficient cardiac pumping or vasospasm, and are closely associated with the risk of hypotension. Detailed analysis of the ascending and descending limbs of the waveform reveals abnormalities such as a slowed ascending limb slope and a deepened descending limb notch, acting like "road warning signs" for microcirculation, indicating microcirculatory disturbances. Poor microcirculation hinders cardiac pumping, preventing effective stroke volume perfusion to the periphery, contributing to hypotension and providing subtle clues for comprehensive diagnosis. Ultimately, the above-mentioned multiple features from different waveforms are organically integrated and embedded with deep insights to comprehensively enhance the risk identification capabilities of the feature library, accurately capture clues of hypotension risks in advance, and seize the initiative for clinical intervention.

[0033] Clinical events occurring during surgery are coded and quantified. For example, intraoperative bleeding events are coded according to the amount of bleeding (mild <500ml, moderate 500-1000ml, severe >1000ml), bleeding rate (slow <50ml / min, medium 50-100ml / min, fast >100ml / min), and anatomical risk classification of bleeding sites (such as high-risk areas such as the brain, heart, and large blood vessels, and low-risk areas such as the limbs and body surface); anesthesia accidents are coded according to type (such as excessive anesthesia leading to respiratory and circulatory depression, allergic reaction, and malignant hyperthermia) and severity (based on the amplitude of vital sign fluctuations and the intensity of intervention measures); arrhythmia events are coded according to the type of arrhythmia (frequency and morphological complexity of sinus, atrial, and ventricular premature beats, type and heart rate range of tachycardia or bradycardia, degree and location of conduction block), duration, and degree of impact on hemodynamics. Clinical event characteristics are integrated into the feature library, correlating them with changes in physiological indicators, accurately depicting the immediate and long-term impact of sudden events during the surgical process on the risk of hypotension, and improving the clinical practicality of the feature library and the accuracy of risk prediction.

[0034] Based on the principle of stratified sampling, the original feature library was divided into training and validation sets at a ratio of 70% to 30%. This ensured a balanced distribution of samples across age groups and complication groups in the training and validation sets, maintaining data consistency. For example, in the subset of elderly hypertensive surgery patients, 70% of the samples were used in the training set to enable the model to learn the physiological characteristics and hypotension patterns of this population, while 30% were used in the validation set to assess the model's generalization ability. This prevented the model from overfitting to specific samples and ensured the accuracy and robustness of the model's predictions of hypotension risk across different age groups and complication types, laying a scientific data framework for precise model training and validation. Machine learning algorithms (such as logistic regression, decision trees, and support vector machines) or deep learning models (such as multilayer perceptrons and convolutional neural networks) were selected for predictions on the validation set. Logistic regression uses a sigmoid function to linearly combine features and output a hypotension probability value between 0 and 1. Decision trees use feature splitting rules to construct decision paths and determine the probability of hypotension categories. Support vector machines find the optimal hyperplane in the feature space to partition the hypotension risk region and output predictions. Deep learning models use multi-layer neuron feature transformation and activation function processing to output a hypotension prediction score or probability.

[0035] In-depth research into the causes of hypotension and the development of precise early warning models requires fully exploiting the physiological and pathological information hidden behind multiple waveform features. First, we analyze hypotension and its causes. Focusing on the ECG waveform, increased P wave dispersion reflects impaired atrial conduction consistency. This disrupts the initial cardiac electrical activity, which can lead to uncoordinated atrial contraction, impairing subsequent ventricular filling and reducing stroke volume (SV), potentially contributing to hypotension. For example, in arrhythmias such as atrial fibrillation, the P wave morphology is extremely disorganized, preventing the atria from effectively assisting the ventricles in pumping blood, leading to a downward trend in blood pressure. A widened or notched QRS complex indicates an increased risk of myocardial block. This acts as a blockage in the heart's electrical circuit, preventing efficient and synchronized myocardial contractions. This significantly impairs pumping function and leads to a sharp decrease in stroke volume (SV), directly inducing hypotension. For example, in patients with myocardial infarction, conduction is disrupted in the myocardium surrounding the infarcted area, often resulting in abnormal QRS complexes and a consequent significant decrease in blood pressure. ST segment depression or elevation accurately pinpoints the duration, extent, and severity of myocardial ischemia. Myocardial ischemia will lead to myocardial contractile weakness and a decrease in stroke volume. Long-term ischemia will also cause myocardial remodeling, further weakening cardiac function and promoting hypotension. When patients with coronary heart disease suffer from angina pectoris, dynamic changes in the ST segment are accompanied by a decrease in blood pressure. Tall, inverted, and flat T waves reveal abnormal myocardial repolarization, disrupting cardiac electrophysiological stability. The coordination of cardiac rhythm and contraction is disrupted, the stroke volume fluctuates, and blood pressure cannot be maintained stably. When electrolyte disorders affect myocardial repolarization, the risk of hypotension increases significantly.

[0036] At the arterial pressure waveform level: a steep decline in the slope of systolic, diastolic, and mean arterial pressure indicates a serious imbalance in the coordination between vascular elasticity and cardiac contraction. Vascular aging and arteriosclerosis reduce elasticity, preventing proper dilation during cardiac contraction. The heart must overcome significant resistance to pump blood, making it difficult to maintain a stable stroke volume, leading to hypotension. Elderly individuals often experience this type of blood pressure fluctuation accompanied by hypotension due to vascular degeneration. Once the dynamic balance between blood volume and vascular tone is disrupted, pulse pressure variation fluctuates abnormally. Sudden blood volume loss due to massive blood loss or dehydration, or excessive vasodilation (such as vasoplegia caused by sepsis), creates significant uncertainty in cardiac stroke output, dramatically increasing the risk of hypotension. Abnormal interactions between peripheral vascular resistance regulation and arterial compliance are reflected in variations in the depth of the dicrotic notch. When systemic vascular resistance (SVR) changes due to peripheral vasospasm (such as Raynaud's disease) or vasodilatory dysfunction, cardiac afterload increases or destabilizes, interfering with stroke volume output, disrupting blood pressure stability, and inducing hypotension.

[0037] At the photoplethysmography level: prolonged pulse wave conduction time or violent fluctuations mean that vascular elasticity is declining and peripheral resistance is unbalanced. This is common in long-term uncontrolled hypertension leading to vascular sclerosis, or microvascular lesions caused by diabetes, which hinder the heart's pumping, impaired stroke volume, and highlight the risk of hypotension. Low or frequently fluctuating perfusion indexes indicate unstable blood supply to peripheral tissues. This may be due to insufficient heart pumping, or it may be due to local vascular lesions (such as peripheral atherosclerosis occlusive disease) that hinder blood flow, making poor tissue perfusion and the risk of hypotension inseparable. Abnormalities such as a slowed slope of the rising branch of the waveform and a deepened notch of the descending branch reveal microcirculatory disorders. For example, in the early stages of shock, microcirculation ischemia makes it difficult for the heart to effectively perfuse the periphery, and the stroke volume is difficult to function, which leads to hypotension.

[0038] Next, to accurately predict the risk of hypotension, the system selects multiple waveform features closely related to hypotension, such as key parameters in the ECG waveform, various arterial pressure waveform indicators, and photoplethysmography features. These features are linearly combined and then fed into a sigmoid function, which cleverly transforms the probability of hypotension between 0 and 1. This probability value acts as a precise "blood pressure risk indicator," intuitively reflecting the likelihood of hypotension in each sample and providing a quantitative basis for subsequent judgments. Based on feature splitting rules mined from extensive clinical data, a methodical decision path is constructed. Each node is a decision branch based on a key feature, progressing layer by layer to ultimately accurately determine the probability of hypotension. For example, a first-level classification is performed based on whether the QRS complex in the ECG waveform is abnormal. This is further refined by combining the pulse pressure variability of the arterial pressure waveform, providing a reliable reference for hypotension risk stratification. The prediction results also include the prediction category (whether hypotension occurs or not), probability value and prediction confidence indicators (such as prediction probability standard deviation and confidence interval). The multi-dimensional results provide a rich data source for subsequent identification information generation, accurately quantify the hypotension risk status of the validation set samples, drive model performance evaluation and optimization iteration, and improve the reliability and effectiveness of the hypotension warning model.

[0039] The prediction results of the validation set were compared with the true labels (hypotension events were strictly defined based on intraoperative blood pressure monitoring). A confusion matrix was constructed to count the number of true positive (TP, hypotension predicted and actually occurred), false positive (FP, hypotension predicted but not actually occurred), true negative (TN, no hypotension predicted and not actually occurred), and false negative (FN, no hypotension predicted and not actually occurred) samples. The accuracy ((TP+TN) / (TP+FP+TN+FN)) was calculated to measure the overall prediction accuracy, the recall (TP / (TP+FN)) to assess the model's ability to identify hypotension when it actually occurred, and the F1 value (2*accuracy*recall / (accuracy+recall)) to comprehensively consider the balance between accuracy and recall. The receiver operating characteristic (ROC) curve was plotted, and the area under the curve (AUC) was used to quantify the model's ability to distinguish hypotensive from normotensive samples (AUCs closer to 1 are better, and >0.7 is clinically valuable). A multi-index evaluation system accurately quantifies the model's predictive performance, identifies model weaknesses, calibrates the accuracy of identification information generation, guides model optimization, and ensures the reliability and clinical guidance of hypotension identification information.

[0040] If the model meets preset performance standards on the validation set (e.g., accuracy >80%, recall >70%, AUC >0.75), a prediction confidence threshold (e.g., a probability >0.6 is considered a high-confidence prediction) is used to determine the identification information. First, the performance requirements for the model on the validation set were established, clarifying quantitative metrics such as accuracy, recall, and AUC. This established a scientific and reasonable basis for judging whether the model is acceptable, ensuring that the model possesses a certain level of reliability and validity. Only when these preset performance standards are met can the subsequent generation of model-based identification information be more credible. Prediction confidence thresholds were then introduced to further subdivide the predictions based on probability values. For example, using the "probability >0.6 is considered a high-confidence prediction" as an example, different confidence intervals correspond to different identification information: samples with a high-confidence prediction of hypotension are labeled "strongly positive," meaning that such samples are generally considered to be highly likely to develop hypotension. This identification allows medical staff to quickly focus on these patients and prioritize them, as these patients are at particularly high risk of developing hypotension. Samples with low-confidence hypotension predictions and high-confidence normotension predictions are labeled "weak positive," a relatively vague definition that contains potential information. A low-confidence hypotension prediction indicates that the sample has a certain tendency towards hypotension but is uncertain, while a high-confidence normotension prediction indicates a high probability of normal blood pressure but still retains a small degree of doubt. Combined, they are labeled "weak positive," prompting medical staff to pay appropriate attention and require continued observation or further investigation to prevent potential risks. High-confidence normotension samples are labeled "negative," which clearly informs medical staff that these samples have almost no risk of hypotension and can be handled routinely without special attention.

[0041] Finally, the practical value of these identification information is emphasized. It accurately characterizes the user's risk level of hypotension. Medical staff can quickly identify high-risk users based on the labels, and then formulate targeted tiered monitoring and intervention strategies to optimize the perioperative hypotension risk management process, ultimately improving medical quality and patient surgical safety. The entire logical chain is complete, and each link works closely together to achieve accurate and efficient medical decision support.

[0042] In another implementation, a deep neural network (DNN) and random forest (RF) fusion model architecture is selected based on the specific characteristics of the surgical hypotension warning task. The DNN automatically exploits deep nonlinear relationships between physiological features, while the RF ensemble learning advantage reduces model variance and improves generalization. The DNN has four hidden layers, with the number of neurons in each layer determined based on the input feature dimension and sample size (e.g., 128 in the first layer, decreasing to 32 in the last layer). Reinforced linear unit (ReLU) activation functions are used to prevent vanishing gradients. The RF constructs a forest of 100 decision trees, with node splitting based on minimizing the Gini index. This ensures a reasonable initial model architecture and balances learning and generalization performance, laying the foundation for fusion training. 70% of the labeled sample set is allocated as a training set, and 30% as a validation set. The training set samples are combined with surgical information (procedure type code, duration, urgency), physiological activity (heart rate, blood pressure waveform parameters, respiratory rhythm characteristics), physiological status (medical history code, laboratory indicators, organ function scores), and hypotension identification into the model. During training, the DNN forward propagates the predicted probability, measuring the difference from the true label using the cross-entropy loss function. Each decision tree in the RF splits and grows predictions based on the features, and the results are aggregated to obtain the voting results. The Adam optimizer is used to update the DNN parameters, and the RF is optimized based on out-of-bag data evaluation. Multiple iterations are performed until the validation set loss converges and performance metrics stabilize (e.g., precision 85%, recall 80%, F1 value 0.82). The model accurately learns the association between sample features and hypotension. Later in training, the DNN and RF predictions are fused. Fusion strategies include weighted averaging (weighting based on validation set performance) or cascade combination (DNN outputs are used as RF input features). After fusion, the overall parameters of the model are fine-tuned. In case of overfitting, early stopping is used to pause training based on validation set performance monitoring, L1 / L2 regularization is used to constrain DNN weights, and RF pruning is used to reduce complexity. This improves model generalization and robustness, optimizes prediction performance to meet clinical application standards, and generates an accurate and reliable target hypotension early warning model, helping medical staff to accurately prevent and control the risk of intraoperative hypotension, improving surgical quality and patient safety.

[0043] S102 : Processing the target user's surgical information within a preset time period to generate an interval prediction subtask and an intraoperative prediction subtask.

[0044] In one embodiment, the historical operation time information of the target user is comprehensively collected from the surgical database and physiological monitoring database of the hospital information system, including the precise start and end time of the operation, the duration of the operation, the type of operation (such as general surgery, orthopedics, neurosurgery, etc.), the urgency of the operation (elective, limited-time, emergency surgery), and the anesthesia method used (general anesthesia, spinal anesthesia, regional block anesthesia, etc.). At the same time, the occurrence records of hypotension in historical operations are collected, covering key details such as the start time, duration, lowest blood pressure value, intervention measures taken and their effect evaluation of the hypotensive attack, to construct a multi-dimensional historical surgery and hypotension data set, laying a solid data foundation for subsequent in-depth analysis. Use data analysis software to conduct in-depth mining of the collected data. Accurately calculate the time relationship between hypotension events in each operation and each key stage of the operation (such as anesthesia induction period, peak surgical operation period, abdominal closure and suturing period), and determine the distribution characteristics of the occurrence period of hypotension during the operation; count the frequency and severity of hypotension under different types of operations and anesthesia methods (based on the degree of blood pressure reduction and duration of classification); analyze the common physiological status of users when hypotension occurs (such as preoperative basal blood pressure level, heart rate variability, blood volume status, etc.), construct a historical surgical hypotension association information set, and accurately present the intrinsic connection and law between surgery and hypotension.

[0045] The target user's surgical records for the past five years were reviewed, covering eight surgical records. Among them, there were three general surgery cholecystectomies and five orthopedic joint replacements. For each operation, the operation duration (e.g., the average duration of cholecystectomy is 120 minutes, and that of joint replacement is about 180 minutes), the amount of bleeding (the average amount of bleeding for cholecystectomy is 200 ml, and that for joint replacement fluctuates between 400-800 ml depending on the complexity of the operation), the anesthesia method (general anesthesia, spinal anesthesia, etc., cholecystectomy mostly uses general anesthesia, and joint replacement partially uses spinal anesthesia combined with general anesthesia), the time period of intraoperative hypotension (hypotension is prone to occur when the gallbladder artery is handled during cholecystectomy, and the risk of hypotension is high 10-20 minutes after prosthesis implantation in joint replacement) and corresponding intervention measures (cholecystectomy should be treated with 500-1000 ml of fluid and ephedrine should be used to increase blood pressure as needed; joint replacement should flexibly adjust the amount of fluid replacement and the dose and drip rate of vasoactive drugs such as dopamine according to the amount of blood loss and blood pressure changes) and other key historical surgical information were accurately extracted. In-depth analysis of the dynamic changes of physiological indicators, accurate calculation of the heart rate variation coefficient (taking a joint replacement surgery as an example, the standard deviation of the heart rate throughout the surgery was 10 beats / minute, the mean was 70 beats / minute, and the variation coefficient was 0.14, showing the fluctuation characteristics of cardiac stress regulation), the mean arterial pressure drop slope (such as the MAP dropped by 15mmHg within 3 minutes in a specific period of time during cholecystectomy, the slope was -5mmHg / min, and the blood pressure sudden change trend was intuitively presented), and accurate quantification of the dynamic change characteristics of physiological indicators during surgery; scoring based on the complexity of the surgical operation (joint replacement surgery involves precise bone cutting and prosthesis adaptation, and the operation complexity is high, with a score of 8 points; cholecystectomy is relatively simple, with a score of 4 points), surgery The density score of vascular nerve distribution in each region (the score for rich vascular nerve distribution in the joint replacement surgery region is 7 points, and the score for relatively concentrated vascular nerve distribution in the cholecystectomy region is 5 points) is used to scientifically quantify surgical operation factors; the frequency of anesthetic drug dosage adjustment (for example, in a general anesthesia joint replacement surgery, the dose of propofol and remifentanil is fine-tuned every 40 minutes based on the user's vital signs and the progress of the surgery) and the fluctuation range of anesthesia depth (measured by BIS value, the BIS value of a certain surgery fluctuates between 40-60 with a standard deviation of 5, accurately reflecting the dynamics of anesthesia depth regulation) are accurately calculated; key indicators of anesthesia management factors are deeply mined, and a detailed historical surgical hypotension-related information set is comprehensively constructed to lay a solid data foundation for subsequent analysis.

[0046] Taking coronary artery bypass grafting as an example, the duration of each key stage is carefully planned according to the standard surgical process (the scheduled time for establishing extracorporeal circulation is 35 minutes, the vascular anastomosis stage is about 100 minutes, and the heart beat and subsequent monitoring is 60 minutes), the blood loss is scientifically estimated (it is estimated to be 1000-1500 ml according to the user's cardiac function classification, degree of coronary artery lesions, and vascular calcification), the anesthesia depth requirement is accurately predicted (the anesthesia depth maintenance range is set according to the difference in stimulation intensity at each stage of the operation, such as BIS value 45-55 during skin incision and thoracotomy, BIS 40-50 during vascular anastomosis) and customized fluid management strategy (based on preoperative The ratio, rate, and total amount of crystalloid and colloid infusion are planned based on blood volume status and expected blood loss. If the blood volume is normal before surgery, the first dose of crystalloid solution is 1000 ml, infused at a constant rate over 30 minutes. Colloid solution is supplemented in a timely manner according to intraoperative bleeding to maintain blood volume stability. At the same time, key basic physiological indicators (heart rate 72 beats / minute, blood pressure 125 / 75 mmHg, blood oxygen saturation 97%) and core laboratory indicators (hemoglobin 125 g / L, blood glucose 5.5 mmol / L, coagulation function indicators PT 12 seconds, APTT 32 seconds, platelet count 220 × 10 9 / L), deeply integrates and generates accurate real-time surgical preset information, firmly anchoring the scientific baseline framework for intraoperative dynamic monitoring, accurate comparison and efficient risk warning, and comprehensively ensuring the orderly progress of surgery.

[0047] During the whole operation, data is collected collaboratively by multiple devices such as high-end bedside monitors and advanced hemodynamic monitors. Arterial blood pressure waveforms are collected at a high frequency per second to accurately calculate real-time blood pressure values (systolic pressure, diastolic pressure, mean arterial pressure accurate to 1mmHg) and heart rate (real-time capture of heart rate fluctuations to an accuracy of 1 beat / minute); continuous monitoring of electrocardiograms to deeply analyze heart rate variability (such as calculating SDNN to be 32ms and RMSSD to be 16ms, accurately insight into the autonomic nervous function of the heart), subtle changes in ST segment morphology (ST segment depression or elevation amplitude accurate to 0.05mV, duration accurately recorded to 1 second, sensitively identifying early signals of myocardial ischemia); photoplethysmography to high-frequency monitoring of pulse wave conduction time (accurately measured to 1ms, such as an average of 122ms), perfusion index (real-time monitoring of fluctuations to 0.1, such as an average of 4.6); respiratory monitors to record respiratory activity with high precision. Frequency (real-time monitoring to 0.1 times / minute, average 15.5 times / minute), tidal volume (accurate measurement to 10ml, average 480ml), end-tidal carbon dioxide partial pressure (real-time monitoring to 0.1mmHg, average 37.5mmHg); transpulmonary thermodilution technology accurately measures core parameters such as cardiac output (accurate to 0.1L / min, such as average 5.2L / min), stroke volume variability (accurate measurement to 1%, such as average 11%), and dynamically updates the data matrix every 3 minutes to deeply construct a dynamic, accurate, and real-time intraoperative physiological feature matrix, laying a solid data foundation for real-time and accurate assessment of the user's physiological reserve, surgical tolerance, and efficient warning of hypotension risks during surgery, thereby comprehensively improving the safety and effectiveness of surgery.

[0048] The innovative architecture of the deep belief network (DBN) and support vector regression (SVR) is used. DBN automatically mines the high-dimensional abstract representation and deep implicit relationship of physiological features with its powerful unsupervised learning ability, while SVR accurately calibrates the predicted value and provides reliable confidence intervals with its excellent small sample learning and generalization performance. By inputting multi-dimensional features of historical surgeries (such as the long-term trend of heart rate variability coefficient, the historical mean of surgical operation complexity, and the statistical characteristics of the frequency of anesthetic drug dose adjustments) and real-time intraoperative physiological parameters (real-time heart rate variability index, real-time operation complexity score of the current surgical stage, and real-time fluctuation value of anesthesia depth), the model outputs the probability of hypotension through complex calculations (such as a predicted probability of 18% at the current surgical node), accurately estimates the critical period of hypotension (the risk will increase significantly within the next 8 minutes), deeply analyzes and classifies potential causes (such as relative blood volume deficiency, determined by the real-time SVV rise and the difference between estimated blood loss and actual blood loss; or myocardial function suppression, which is diagnosed collaboratively by real-time ST segment dynamics and myocardial contractility indicators), and intelligently generates preliminary intervention recommendations (if blood volume is insufficient, quickly replenish colloidal fluid at a standard of 15ml / kg and monitor CVP dynamics; for myocardial suppression, accurately select drugs based on changes in heart rate and blood pressure, such as the dobutamine micropump maintenance dose adjusted according to cardiac function indicators), thus generating comprehensive and accurate and intelligent surgical hypotension prediction information.

[0049] The task architecture is generated based on information science for accurately predicting surgical hypotension. The interval prediction subtask, guided by evidence-based medicine big data and clinical guidelines, combines the characteristics of the current surgery and the user's recovery trends to accurately predict the probability of recurrent hypotension within the next week, month, and year. (For example, the risk of recurrent hypotension is predicted to be 12% within one week, 18% within one month, and 25% within one year. This allows for the development of personalized postoperative follow-up monitoring frequency, examination package, and rehabilitation intervention intensity tiered plans based on risk levels.) The intraoperative prediction subtask provides detailed, tiered warnings based on key milestones in the surgical process and dynamic changes in physiological indicators. Such as a mild warning 3 minutes before a critical operation (a friendly reminder is pushed to urge the medical team to pay close attention to the trends of blood pressure and related physiological indicators), a moderate warning when blood pressure fluctuations exceed the threshold during the operation (an in-depth risk assessment report is pushed immediately to guide the medical team to quickly evaluate the root cause of the risk within 3 minutes, activate the emergency plan and prepare intervention measures), and an emergency alarm at the moment of hypotension (an accurate and personalized intervention plan is pushed immediately to intelligently guide the medical team to accurately operate according to the plan and dynamically optimize the treatment strategy). The task parameters are dynamically and intelligently optimized according to the real-time intraoperative indicators throughout the process, and the risk threshold is accurately calibrated to build a fully closed-loop, intelligent intraoperative hypotension precision control system, which comprehensively improves the safety and quality of surgery, optimizes the user's prognosis and rehabilitation results, and creates a new era of surgical hypotension management.

[0050] S103 , processing the interval prediction subtask and the intraoperative prediction subtask to generate a preset prediction feature vector.

[0051] In one embodiment, the interval prediction subtask and the intraoperative prediction subtask are processed to generate target feature information. This target feature information includes the real-time inter-operative time interval, the target user's physiological indicator mean, the slope of physiological indicator change, the intraoperative operation duration, and the fluctuation of anesthesia depth. Accurately controlling the risk of hypotension during real-time surgery requires comprehensive and dynamic integration of various key factors. This requires not only focusing on the physiological changes currently occurring in the operating room, but also reviewing the patient's past surgical history to uncover potential risk clues, thereby building a comprehensive hypotension risk assessment system. On the one hand, a focus is placed on real-time dynamic monitoring. From the moment the surgery begins, the real-time fluctuations of various physiological indicators are closely tracked. For example, the patient's heart rate curve is continuously recorded, observing the subtle differences in each beat to identify any abnormal nodes of sudden acceleration or deceleration. Accurately measuring blood pressure requires more than a simple reading; it also requires analyzing the fluctuation trend over a short period of time. For example, the mean blood pressure value and the fluctuation amplitude are calculated every 5 minutes. Through this high-frequency data collection, potential signs of instability in the blood pressure regulation mechanism can be captured. At the same time, pay attention to the dynamic changes of hemoglobin in real time to understand the real-time status of oxygen transport capacity under surgical stress. If the hemoglobin content is found to have a downward trend, it means that the body's oxygen supply capacity may be impaired, which is closely related to the increased risk of hypotension.

[0052] On the other hand, we conducted an in-depth review of the surgical history. When reviewing the target user's past surgical experience, the time interval between surgeries is still an important consideration. The intervals between the user's six surgeries in the past five years were 12 months, 18 months, 9 months, 24 months, 15 months, and 6 months, respectively, with an average of 15 months and a standard deviation of 6 months. But this time we emphasized how to combine this historical information with the current real-time situation: although a longer interval between previous surgeries is theoretically conducive to physiological recovery, if the patient's physiological indicators fluctuate abnormally during the current surgery, even with relatively ample recovery time, it may imply that the difficulty of this operation is far beyond expectations, the body's reserves are difficult to cope with, and the risk of hypotension increases sharply; on the contrary, if the interval between previous surgeries is short and the body is under heavy stress, combined with real-time monitoring during the current surgery, once it is found that the heart rate and blood pressure regulation ability are weak, it can be predicted in advance that the probability of hypotension is greater.

[0053] Secondly, the intraoperative operation details are combined. According to the type of real-time surgery, the intraoperative operation time is carefully counted. For example, if a joint replacement surgery is currently being performed, the time is counted from the moment the skin is incised. Compared with the average duration of similar surgeries, if it is significantly exceeded, it means that the trauma to the body is prolonged, the physiological consumption is aggravated, the risk of internal environment disorder is increased, and the risk of hypotension rises accordingly. At the same time, the fluctuation of anesthesia depth is closely monitored throughout the process, measured by the standard deviation of the BIS value. The BIS value is recorded every 10 minutes and the standard deviation is calculated. Once the fluctuation amplitude is found to increase, it is like a ship that has lost its way in a turbulent river. The cardiovascular regulatory center loses control and the risk of hypotension increases sharply. Finally, by integrating the mean and change slope of the real-time monitored physiological indicators, as well as the retrospective inter-operative time interval value, intraoperative operation time and anesthesia depth fluctuation value and other multi-dimensional information, a real-time dynamic target feature information set is systematically and accurately constructed. This information set acts like a precise navigator, comprehensively and deeply quantifying the intricate relationship between surgical history and real-time physiological characteristics. It provides a strong basis for medical staff to accurately assess and promptly intervene in the risk of hypotension during surgery, ensuring the safe and smooth progress of the operation.

[0054] Real-time intraoperative monitoring collects data regularly based on the progress of the surgery. For example, during coronary artery bypass surgery, the average heart rate before extracorporeal circulation is 78 beats / minute, blood pressure is 130 / 80 mmHg, and the stroke volume variability is 12%. During the circulation, heart rate, blood pressure, central venous pressure, and urine output are recorded every 5 minutes. When the heart rate rises to 90 beats / minute, blood pressure drops to 110 / 65 mmHg, CVP rises by 5 mmHg, and urine output decreases by 20 ml / h, the data are continuously monitored and integrated into dynamic target feature information. The integration of historical surgical experience and real-time physiological data comprehensively reflects the evolution of the physiological state throughout the surgery, deeply explores the impact of the dynamic interaction between surgical factors and physiological indicators on the risk of hypotension, and provides core data support for accurate prediction.

[0055] The target feature information is processed to generate the first feature association value and the second feature association value, wherein the first feature association value is used to characterize the correlation between the operation interval time pattern and the change of physiological characteristics, and the second feature association value is used to characterize the correlation between the surgical characteristics and the risk of hypotension. The first feature association value is calculated using the Pearson correlation coefficient and the mutual information method. For example, when analyzing 30 users of similar surgeries, the Pearson coefficient between the operation interval time and the mean heart rate before surgery was -0.4, and the short interval had a high heart rate, indicating that frequent surgery inhibited the regulation of the cardiac autonomic nervous system and affected the heart rate reserve; the mutual information between the operation interval and the slope of intraoperative blood pressure change was 0.35 bits, and the short interval caused the body to be stressed and the blood pressure fluctuated greatly. From the time dimension, the impact of surgical rhythm on physiological function remodeling and stress response is quantified, and the synergistic law of the dynamic change of the operation interval time pattern and physiological characteristics is accurately located, providing key clues for tracing the source of hypotension risk, and assisting the clinician in pre-adjusting the user's physiological state according to the surgical plan and optimizing the choice of surgical timing. Specifically, the calculation formula for generating the first feature association value is: Among them, x i is the sample value of the surgery interval duration, is its mean; y i is the physiological characteristic value (such as mean heart rate, mean blood pressure, etc.), is the mean of the physiological characteristic, and n is the number of samples. r ranges from -1 to 1, with an absolute value close to 1 indicating strong correlation and 0 indicating no correlation.

[0056] The second characteristic correlation value was calculated, and a surgical feature matrix (including surgical type, duration, complexity score, anesthesia method, estimated and actual blood loss) and a hypotension occurrence matrix (0-1 coding and severity grading) were constructed, and analyzed using logistic regression and decision tree models. For example, the risk of hypotension in cardiac surgery is 3 times higher than that in general surgery (OR=3.0); the risk increases by 50% when the operation lasts more than 3 hours (RR=1.5); the risk of complex spinal surgery (complexity score 9) is 2 times that of simple surgery; general anesthesia causes 40% more hypotension than local anesthesia (RR=1.4); the risk of blood loss exceeding 800ml increases dramatically by 80% (RR=1.8). Multi-factor collaborative analysis of the causal relationship between the inherent attributes of surgery and hypotension, and the construction of a surgical risk stratification model, can help with accurate preoperative risk assessment, optimize surgical plans and resource allocation, and improve the effectiveness of intraoperative hypotension risk management and the level of precise prevention and control. Assume that the surgical feature vector X=(x1, x2, ..., x n )(including coding values such as surgery type, duration, and anesthesia method), the second feature association value β i is the regression coefficient. OR = e βt The effect of the t-th surgical feature on the risk of hypotension was measured, with an OR > 1 increasing the risk and an OR < 1 decreasing the risk.

[0057] The first feature association value and the second feature association are processed to generate a preset prediction feature vector. The correlation between the operation interval time pattern and the physiological characteristics change is calculated using 60 surgical users as an example. The average operation interval time (month) Standard deviation S x =4; mean heart rate times / minute, standard deviation S y =7. Some sample values: User 1 (x1=8, y1=82), User 2 (x2=16, t2=72), etc. Calculated according to the Pearson correlation coefficient formula: The calculated r=-0.45 indicates a moderate negative correlation between the duration of the surgical interval and the mean heart rate. Users undergoing short-interval surgeries have higher mean heart rates, which are attributed to insufficient recovery and frequent cardiac stress adjustments, impacting cardiac function reserve and intraoperative blood pressure stability. This serves as a key indicator embedded in the surgical rhythm dimension for hypotension risk assessment, helping to predict cardiac compensation capacity and hypotension propensity under surgical stress.

[0058] The correlation between surgical characteristics and the risk of hypotension was analyzed. A logistic regression model was constructed with the type of surgery (cardiac surgery x1 = 1, general surgery x1 = 0), surgery duration (hours) x2, and anesthesia method (general anesthesia x3 = 1, local anesthesia x3 = 0) as independent variables and the occurrence of hypotension (y = 1 or y = 0) as the dependent variable. 0.9 =2.46, that is, the risk of hypotension in cardiac surgery is 2.46 times that in general surgery; for every hour increase in operation time, OR = e 0.06 =1.06, the risk increased by about 6%; general anesthesia vs. local anesthesia, OR = e 0.5 =1.65, a 65% increase in risk. This multi-factor collaborative quantification of the contribution of surgical characteristics to hypotension risk provides a quantitative basis for surgical planning, anesthesia selection, and risk prevention, accurately weights key risk factors, and improves the targeting and accuracy of hypotension risk management.

[0059] A correlation coefficient matrix A between surgical intervals and physiological characteristics was constructed, including heart rate -0.45, blood pressure 0.32 (calculated using the same method as heart rate), hemoglobin -0.28 (similarly calculated to negative correlation, short intervals deplete reserve), blood oxygen saturation 0.18 (weak positive correlation, reflecting that oxygen supply and demand vary with intervals), and respiratory rate 0.09 (low correlation): A = [-0.45 0.32 - 0.28 0.18 0.09] T ; Surgery-hypotension risk correlation coefficient matrix B, surgery type (cardiac surgery column is subdivided by complexity), duration, anesthesia method, blood loss: For example, complex cardiac surgery 3.20 is a high-risk indicator, the duration column is quantified according to the risk increase, the anesthesia and blood loss OR values are assigned according to the risk effect, and the matrix comprehensively quantifies the risk weights of surgical factors, which is a feature vector fusion of surgical risk characteristics. The above-mentioned construction contains a matrix of associated values, physiological statistics, and an element vector of surgical intervals, and is weighted according to the importance of the features. The weight of the surgical type (dominated by high-risk cardiac surgery), the heart rate correlation value (strong core physiological correlation), the blood pressure mean, the duration of the operation (key risk factor), etc. The high-dimensional vector integrates multiple key information, accurately quantifies the overall picture of surgical physiological risks, and inputs core data drive into the intelligent early warning model, efficiently identifies the risk of hypotension, optimizes the blood pressure management decision chain throughout the operation, and improves the safety quality of the operation and the level of user prognosis.

[0060] S104: Process the physiological motion parameter information of the target user to generate a blood pressure limit factor of the target user.

[0061] In one embodiment, a high-precision respiratory monitor is used to collect the target user's respiratory waveform at a sampling rate of 100Hz. Features such as the inspiratory phase slope, expiratory phase slope, respiratory cycle duration, and waveform amplitude are extracted. For example, in a certain operation, the average inspiratory phase slope is 0.35V / s, reflecting the speed and strength of inspiration; the expiratory phase slope is -0.28V / s, reflecting the characteristics of the exhalation process; the respiratory cycle is 4.5s and the amplitude fluctuates by 0.8V, laying the foundation for evaluating respiratory function and oxygen supply. From the waveform morphology, normal and abnormal waveforms (such as delayed descending limb and concave changes in obstructive ventilation waveforms) are distinguished based on the roundness and symmetry of the peaks and troughs, and a dynamic respiratory waveform feature set is constructed to accurately quantify the dynamic changes in respiratory mechanics and rhythm. A 12-lead electrocardiogram device is used to collect ECG signals in real time, and the dynamic characteristics of R wave peak spacing (RR interval), ST segment morphology changes, T wave morphology and amplitude are extracted through signal processing algorithms. Heart rate variability is calculated by calculating the standard deviation of the RR interval. For example, a standard deviation of 25ms during surgery indicates cardiac autonomic nervous system dysfunction. A horizontal or downward-sloping ST segment depression of 0.1mV for 80ms indicates myocardial ischemia. T-wave elevation, inversion, or flattening reflects abnormal myocardial repolarization. Multi-lead comprehensive analysis captures the spatiotemporal dynamics of the ECG vector, constructing an ECG signal feature set and deeply analyzing the potential impact of cardiac electrophysiological activity on circulatory stability and blood pressure.

[0062] Blood oxygen saturation (SpO2) is continuously monitored using a fingertip pulse oximeter, with data collected every second to record its mean, standard deviation, and fluctuation frequency. For example, during surgery, the mean SpO2 was 96%, with a standard deviation of 1.5% and a fluctuation of 5 times per minute. During a hypoxic event, the lowest value recorded was 88% for 2 minutes and a recovery time of 4 minutes, which is associated with respiratory and circulatory dysfunction. By analyzing the SpO2 decline rate (e.g., 2% / min), the recovery slope (e.g., 3% / min), and its synchronization with heart rate and respiration, a dynamic feature set of blood oxygen saturation is constructed. From the perspective of oxygen delivery, the causal relationship between it and blood pressure regulation imbalance is explored, adding key parameters to the multidimensional assessment of hypotension risk. A sidestream carbon dioxide monitor is used to measure the end-tidal carbon dioxide partial pressure (PETCO2) in real time to obtain the partial pressure mean, rising or falling slope, and waveform morphology characteristics. During surgery, the average PETCO2 value is 38 mmHg, and an increase of 0.3 mmHg / min may indicate hypoventilation. Alveolar ventilation uniformity can be determined based on waveform morphology (e.g., a steep inspiratory phase and a plateau expiratory phase are normal, while an abnormal waveform suggests lung disease or circulatory disorders). By analyzing PETCO2 dynamics in conjunction with changes in respiratory rate and tidal volume, a dynamic partial pressure feature set is constructed to reflect the matching status of pulmonary ventilation and perfusion and the efficiency of respiratory metabolite clearance. This embeds key lung function indicators into hypotension warnings, expanding the dimension of circulatory-respiratory interaction risk assessment.

[0063] Dynamic respiratory waveform features, dynamic ECG signal features, dynamic blood oxygen saturation features, and dynamic end-tidal carbon dioxide partial pressure dynamic features are processed to generate characteristic parameter information for different respiratory cycles. The respiratory waveform features accurately identify the start and end points of the respiratory cycle, and for each cycle, the inspiratory time fraction (Ti / Ttot), expiratory time fraction (Te / Ttot), inspiration-expiratory ratio (Ti / Te), coefficient of variation of tidal volume (CVVT), and minute ventilation (MV) are calculated. For example, in a certain cycle, Ti / Ttot = 0.42, Te / Ttot = 0.58, Ti / Te = 0.72, CVVT = 12%, and MV = 8 L / min. Multi-parameter respiratory function is quantified from respiratory phase, volume stability, and ventilation efficiency. Simultaneously, the intra-cycle mean PETCO2, slope coefficient of variation, and mean phase difference (5°) between SpO2 and PETCO2 are extracted to reflect the dynamic coordination of gas exchange and the stability of respiratory efficiency. This enriches the set of respiratory cycle characteristic parameters, deeply exploring the impact of multidimensional fluctuations in respiratory function on circulatory homeostasis and providing detailed respiratory physiological insights into blood pressure limiting factors. Sequence analysis of characteristic parameters of continuous respiratory cycles was performed, and the mean, standard deviation, coefficient of variation, and autocorrelation coefficient were calculated. For example, the mean of Ti / Ttot over 20 cycles was 0.43, the standard deviation was 0.03, the coefficient of variation was 7%, and the autocorrelation coefficient was 0.35 (lagged one cycle), indicating a cyclical trend. The risk of respiratory decompensation was identified based on the dynamic rate of change of parameters (e.g., MV change rate of 15% / 5 minutes) and the frequency of abnormal fluctuations (e.g., CVVT exceeding the threshold by 20% three times / 10 minutes). This allowed the integration of dynamic respiratory evolution information into blood pressure limiting factors, accurately characterizing the boundary conditions for blood pressure fluctuations caused by respiratory instability, and improving the accuracy of the analysis of blood pressure regulatory factors.

[0064] The characteristic parameter information of different respiratory cycles is processed to generate a blood pressure limiting factor (BLF) for the target patient. A multivariate linear regression model is constructed to calculate the blood pressure limiting factor (BLF). The independent variables are the characteristic parameter set of the respiratory cycle (Ti / Ttot, Te / Ttot, Ti / Te, CVVT, MV, PETCO2 slope coefficient of variation, etc.) and key ECG and blood oxygen saturation characteristics (heart rate variability, ST-segment depression integral, SpO2 fluctuation frequency). The model is trained and optimized based on a large number of surgical sample data, such as a sample containing intraoperative physiological data and blood pressure stabilization interval records of 200 different surgical patients. The model is fitted to obtain the fluctuation frequency in Equation 2. The coefficients are calibrated using the least squares method to reflect the strength of each variable's effect on blood pressure limiting. For example, a value of 0.25 (corresponding to MV) indicates that each liter / minute of minute ventilation contributes 0.25 units to the BLF. This quantifies how respiratory, ECG, and oxygen supply factors synergistically limit the range of blood pressure fluctuations, generating a precise individual blood pressure limiting factor and providing a quantitative decision-making basis for intraoperative blood pressure safety management.

[0065] The leave-one-out cross-validation method was used to evaluate the generalization ability of the model and calculate the mean square error (MSE), mean absolute error (MAE) and coefficient of determination (R2 ). For example, MSE=0.85, MAE=0.72, R 2 =0.82 The model has high accuracy and good generalization. The model is continuously optimized based on new sample data and clinical feedback. In cases where model predictions deviate significantly (such as rare respiratory muscle diseases), parameters are adjusted or new features (such as respiratory muscle strength indicators) are incorporated to optimize BLF calculation. Dynamic validation and optimization ensure that the blood pressure limiting factor accurately adapts to individual physiological and pathological variations and complex surgical scenarios, improving the reliability and accuracy of intraoperative blood pressure management and providing a strong defense for surgical safety.

[0066] S105: Process the physiological status information of the target user to generate a dynamic blood pressure influencing factor of the target user.

[0067] In one embodiment, the target user's physiological state information is processed to generate waveform parameter characteristic information, respiratory parameter characteristic information, and cardiac motion parameter information. The waveform parameter characteristic information is processed to generate the systolic pressure coefficient of variation, the diastolic pressure coefficient of variation, the mean arterial pressure slope coefficient of variation, the heart rate variability low-frequency to high-frequency power ratio, the ST segment depression or elevation amplitude integral, the QRS complex duration coefficient of variation, the pulse wave transmission time coefficient of variation, the perfusion index coefficient of variation, the end-tidal carbon dioxide partial pressure coefficient of variation, the carbon dioxide waveform inspiratory phase slope coefficient of variation, and the expiratory phase slope coefficient of variation. The respiratory parameter characteristic information is processed to generate the respiratory rate standard deviation, the respiratory depth coefficient of variation, and the inspiratory time to expiratory time ratio coefficient of variation. The cardiac motion parameter information is processed to generate the heart rate standard deviation, the arrhythmia score, and the myocardial contractility index.

[0068] Specifically, the coefficient of variation (CV) of systolic blood pressure was extracted from the arterial blood pressure waveform. SBP ) and the coefficient of variation of diastolic blood pressure (CV DBP For example, during a surgical operation, the systolic blood pressure fluctuated between 110-135 mmHg within 30 minutes, with an average of 122 mmHg. Diastolic blood pressure 65-80 mmHg, average 72 mmHg, calculated Quantify the short-term fluctuation of blood pressure. MAP-slope ) Calculation: Assume that the MAP values for the next 5 minutes are 90, 92, 95, 93, and 91 mmHg, and calculate the standard deviation and mean ratio according to the linear regression fitting slope sequence to obtain the CV MAP-slope , reflecting the stability of MAP changes, and high variability suggests the risk of abnormal blood pressure regulation.

[0069] The ratio of low-frequency to high-frequency power (LF / HF) of heart rate variability was extracted from the electrocardiogram (lead II). For example, the power of the LF band during surgery was 300ms. 2 , HF band 200ms2 LF / HF=1.5, and the ratio imbalance (such as persistent >2 or <0.5) indicates that the sympathetic-parasympathetic nerve regulation is abnormal and affects the blood pressure rhythm. ST segment depression or elevation amplitude integral: The integral is calculated based on the ST segment deviation 60-80ms after the J point. The ST segment is depressed by 0.1-0.2mV for 5 minutes in a certain period. The integral reflects the impact of the degree of myocardial ischemia on blood pressure. The coefficient of variation of QRS complex duration (CV QRS ) Calculation: For example, if the QRS wave width of 10 heartbeats is 80-100ms and the average is 90ms, Abnormal changes show abnormal myocardial conduction and stable heart pumping and blood pressure.

[0070] The coefficient of variation (CV) of pulse wave transit time extracted from photoplethysmography PPT ) and the coefficient of variation of perfusion index (CV PI If the PPT fluctuates between 120-140ms during surgery, with an average of 130ms, Reflects fluctuations in vascular elasticity and peripheral resistance; PI fluctuates between 3 and 6, with an average of 4.5. Low or high perfusion fluctuations are associated with blood pressure changes. EtCO2 ) and coefficient of variation of slope of inspiratory and expiratory phases CV Insp-slope and CV Exp-slope If EtCO2 fluctuates between 35-42 mmHg and the average is 38 mmHg, The calculation of the slope coefficient of variation is the same as the MAP slope method, which helps to evaluate the impact of respiratory metabolism on circulation and its relationship with blood pressure regulation.

[0071] Standard deviation of respiratory rate (SD RR ) Calculation: If the respiratory rate is 14-18 times / minute for 60 consecutive seconds, the mean is 16 times / minute, and the standard deviation is 1.2 times / minute, SD RR Reflects the stability of respiratory rhythm. High-frequency fluctuations can easily lead to unstable intrathoracic pressure, interfering with venous return and blood pressure. Depth ):Calculated based on the variation of tidal volume or peak inspiratory flow rate. If the tidal volume is 450-550ml and the mean is 500ml, Abnormal fluctuations affect pulmonary ventilation efficiency and oxygenation, and indirectly change blood pressure. I:E ): In a certain period of time, the inhalation is 1.5-2 seconds and the exhalation is 2-2.5 seconds, with an average ratio of 0.75. Abnormal ratios interfere with respiratory mechanical balance, affecting the interaction between intrathoracic pressure and cardiovascular function, causing blood pressure fluctuations.

[0072] Heart rate standard deviation (SD) HR) Statistics: For example, if the heart rate in 10 minutes is 70-85 beats / minute, the average is 78 beats / minute, It reflects the stability of cardiac rhythm, and an increase in the standard deviation indicates the risk of arrhythmia, which affects cardiac output and blood pressure. The Rhythm Arrhythmia Score (RAS) is scored according to the type, frequency, and severity of the arrhythmia. For example, if there are 5 occasional atrial premature beats (1 point / beat) and 2 pairs of ventricular premature beats (3 points / beat), the RAS = 11 points. The impact of cardiac electrophysiological disorders on the amplitude and duration of blood pressure fluctuations is assessed according to the score stratification to guide the priority of intervention. The myocardial contractility index (MSI) is estimated by echocardiography using left ventricular ejection fraction (LVEF), myocardial work parameters, or pressure-volume loop analysis. For example, if the LVEF is 55% and the myocardial work index is 1.2, a comprehensive assessment of myocardial contractile function is performed. Weak contractility leads to decreased cardiac output and increased risk of low blood pressure, providing a core indicator of cardiac function for blood pressure control.

[0073] The waveform parameter characteristic information, the respiratory parameter characteristic information, and the cardiac motion parameter information are processed to generate a dynamic blood pressure impact value. The method includes obtaining a calculation formula for the dynamic blood pressure impact value, which is:

[0074] Among them, ω1, ω2, ω3, ω4, ω5, and ω6 are weight coefficients, CV SBP represents the coefficient of variation of systolic blood pressure, CV DBP represents the coefficient of variation of diastolic blood pressure, CV MAP-slope represents the coefficient of variation of the slope of mean arterial pressure, LF / HF represents the ratio of low-frequency to high-frequency power of heart rate variability, STI represents the integral of the amplitude of ST segment depression or elevation, CV QRS represents the coefficient of variation of QRS complex duration, CV PPT represents the coefficient of variation of pulse wave transit time, CV PI represents the coefficient of variation of perfusion index, CV EtCO2 stands for the coefficient of variation of end-tidal carbon dioxide partial pressure, CV Insp-slope represents the coefficient of variation of the slope of the inspiratory phase of the carbon dioxide waveform, CV Exp-slope represents the coefficient of variation of the expiratory phase slope, SD RR Represents the standard deviation of respiratory rate, CV Depth Respiratory depth coefficient of variation, CV I:E represents the coefficient of variation of the ratio of inspiratory time to expiratory time, SD HRrepresents the heart rate standard deviation, RAS represents the arrhythmia score, and MSI represents the myocardial contractility index. Assuming ω1 = 0.2, ω2 = 0.18, ω3 = 0.15, ω4 = 0.12, ω5 = 0.1, and ω6 = 0.25, weigh the importance of each parameter group's impact on blood pressure. For example, if a user's DBIV = 3.2 (unitless, relative value) after calculating the above parameters, this comprehensive quantification of the combined impact of physiological state fluctuations on blood pressure provides a quantitative basis for risk stratification of individual blood pressure dynamics.

[0075] The dynamic blood pressure impact value is processed to generate the dynamic blood pressure impact factor of the target user. The dynamic blood pressure impact factor (DBIF) is obtained by normalizing the DBIV to 0-1: Assume that the minimum DBIV value is 1.5 and the maximum value is 6.5. When a user's DBIV is 3.2, The system is stratified based on DBIF thresholds: <0.3 low risk, 0.3-0.7 medium risk, and >0.7 high risk. This provides key decision-making indicators for medical staff to accurately identify and stratify the risk of intraoperative hypotension, help optimize blood pressure control strategies during surgery, and improve users' surgical safety.

[0076] S106 , processing the preset prediction feature vector, the target user's blood pressure limit factor, and the target user's dynamic blood pressure influencing factor based on the target hypotension warning model to generate hypotension warning information for the target user.

[0077] In one embodiment, based on the target hypotension warning model, the preset prediction feature vector, the target user's blood pressure limit factor and the target user's dynamic blood pressure influencing factor are subjected to feature fusion processing to generate an initial feature matrix. The preset prediction feature vector includes an index of correlation between the operation interval time pattern and physiological characteristics (such as the correlation value between the operation interval and heart rate -0.45), an index of correlation between the operation characteristics and the risk of hypotension (the OR value corresponding to the operation type cardiac surgery is 2.46), the mean of the physiological index (heart rate normalized value 0.4), the operation interval statistic (mean normalized 0.5), etc.; the blood pressure limit factor is calculated by the respiratory cycle characteristic parameters (inspiratory time proportion 0.42) and the electrocardiogram and blood oxygen characteristics (heart rate variability index value) through the model (such as value 0.32); the dynamic blood pressure influencing factor is generated by weighting according to the waveform parameters (systolic blood pressure variation coefficient 0.066), respiratory parameters (respiratory frequency standard deviation 1.2), and cardiac motion parameters (heart rate standard deviation 0.064) (value 0.45). Each feature outlines the hypotension risk profile from multiple dimensions of surgical history and real-time physiological dynamics, laying the foundation for fusion.

[0078] A fusion strategy combining principal component analysis (PCA) and deep neural networks (DNNs) was used. PCA reduces the dimensionality of high-dimensional features and removes redundancy, retaining principal components with 90% variance explanation. The new features encompass key components of multivariate information, such as the "surgical stress-physiological regulation principal component" that integrates surgical duration and physiological fluctuations. The PCA-processed features were input into a DNN for further abstraction and fusion. The DNN architecture contained three hidden layers (with the number of neurons set to 64, 32, and 16, respectively, depending on the number of features and sample size). Network weights were optimized through forward and backpropagation training, and multivariate features were fused into a matrix form, such as an 8×1 initial feature matrix. The matrix elements condense the hypotension risk characteristics of all surgical factors, providing a high-dimensional, abstract data representation for risk quantification.

[0079] Linear discriminant analysis (LDA) is used to reduce the dimensionality of the initial feature matrix. LDA projects the data based on the principle of maximum inter-class scatter and minimum intra-class scatter, and finds the best projection direction to compress the matrix into one dimension. For example, based on the characteristic distribution of hypotension and normotension samples in a large number of surgical samples, the inter-class and intra-class scatter matrices are constructed, and the projection vector is calculated. Project the matrix M to obtain the scalar value y=w T M=0.42×0.23+0.38×0.18+...+0.29×0.31=0.35, this value is the preliminary estimate of the initial probability value of hypotension for the target user, which realizes the efficient mapping of high-dimensional feature space to one-dimensional risk probability and simplifies the complexity of risk judgment. The initial probability value is calibrated using the Platt scaling method. A logistic regression model is constructed to fit the relationship between the predicted probability and the true label (hypotension or not). The training sample contains 300 surgical data and the corresponding true labels. The calibration parameters a=0.8, b=0.1 are obtained based on sample training, and the probability after calibration is For example, when y=0.35, Provide reliable quantitative basis for clinical decision-making, helping medical staff to accurately weigh the timing and intensity of hypotension intervention.

[0080] Refine the hypotension risk assessment based on real-time surgical preset information. For example, for heart bypass surgery, the preset extracorporeal circulation time is 90 minutes, the expected blood loss is 800 ml, the key operation nodes (vascular anastomosis period) and the anesthesia depth range (BIS value 40-50) are combined with the current initial probability of hypotension of 0.39 and trend analysis (such as the probability increased by 0.05 in the past 10 minutes). If the operation is close to a critical operation and the probability approaches the threshold (such as the set threshold of 0.5), the risk is increased to medium risk; if the intraoperative blood loss is close to the expected value and the probability fluctuates greatly, the risk level is dynamically adjusted, and comprehensive hypotension information is generated according to the risk level. The generated information includes risk level (low, medium, high), predicted hypotension period (medium risk within the next 20 minutes), potential predisposing factors (such as potential blood volume deficiency due to near-estimated blood loss, increased demand for vasoactive drugs due to blood pressure fluctuations associated with myocardial function fluctuations) and intervention recommendations (based on the risk level: low risk, increase monitoring frequency to 5 minutes / time; medium risk, prepare 500ml of colloid solution as a backup and fine-tune the depth of anesthesia; high risk, immediately infuse 300ml of colloid solution and use a dopamine micropump to maintain the dose according to the blood pressure response). It is pushed to medical care through a visual interface or structured report to help accurately prevent and control hypotension risks, optimize surgical management processes, and ensure surgical safety and user prognosis.

[0081] This application is dedicated to accurately predicting the risk of intraoperative hypotension and building a complete closed-loop hypotension early warning system. First, the training sample set is grouped by age and complications, and the physiological characteristics of different groups are mined to build an original feature library. After multiple rounds of iteration, accurate labels are generated for model training. When processing the target user's surgical information, historical data is mined to obtain hypotension-related features, and real-time parameters are combined to generate predictive information and decompose subtasks. The risk weight is quantified by calculating the correlation value, and the blood pressure limiting factor and dynamic blood pressure influencing factor are obtained from the physiological parameter analysis to quantify blood pressure fluctuations. Finally, the fusion vector and factor are used to construct a feature matrix and reduce the dimension. Combined with real-time information calibration, hypotension early warning information containing risk level and other content is generated to provide a basis for clinical decision-making, improve surgical safety and user prognosis quality.

[0082] In one embodiment, Figure 2 As shown, the present application also provides an early warning device for intraoperative hypotension, comprising:

[0083] An acquisition module 201 is configured to acquire surgical information of a target user within a preset time period, physiological motion parameter information of the target user, physiological state information of the target user, a training sample set, and a preset hypotension warning model;

[0084] The processing module 202 is used to pre-process the training sample set to generate a training sample set with identification information, wherein the identification information is used to characterize that the user is in an abnormal hypotension state; based on the training sample set with identification information, the preset hypotension warning model is processed to generate a target hypotension warning model; the surgical information of the target user within a preset time period is processed to generate an interval prediction subtask and an intraoperative prediction subtask; the interval prediction subtask and the intraoperative prediction subtask are processed to generate a preset prediction feature vector; the physiological motion parameter information of the target user is processed to generate a blood pressure limitation factor of the target user; the physiological state information of the target user is processed to generate a dynamic blood pressure influencing factor of the target user; based on the target hypotension warning model, the preset prediction feature vector, the blood pressure limitation factor of the target user and the dynamic blood pressure influencing factor of the target user are processed to generate hypotension warning information of the target user.

[0085] Each embodiment in this application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the early warning method, electronic device, electronic device, and readable storage medium for evaluating intraoperative hypotension, since they are basically similar to the embodiment of the early warning method for intraoperative hypotension described above, the description is relatively simple, and the relevant parts can be referred to the partial description of the embodiment of the early warning method for intraoperative hypotension described above.

Claims

1. A method for early warning of hypotension during surgery, characterized in that: include: Obtaining the target user's surgical information, target user's physiological motion parameter information, and target user's physiological status information within a preset time period; Process the target user's surgical information within a preset time period to generate interval prediction subtasks and intraoperative prediction subtasks; Processing the interval prediction subtask and the intraoperative prediction subtask to generate a preset prediction feature vector; Processing the target user's physiological motion parameter information to generate the target user's blood pressure limit factor; Processing the physiological status information of the target user to generate the dynamic blood pressure influencing factor of the target user; Based on the target hypotension warning model, the preset prediction feature vector, the target user's blood pressure limit factor and the target user's dynamic blood pressure influencing factor are processed to generate hypotension warning information for the target user.

2. The method according to claim 1, wherein The target user's surgical information within the preset time period is processed to generate interval prediction subtasks and intraoperative prediction subtasks, including: Processing the target user's surgical information within a preset time period to generate the target user's historical surgical information and the target user's real-time surgical preset information; Process the target user's historical surgical information to generate historical surgical hypotension related information; Perform feature extraction processing on historical surgical hypotension-related information to generate historical surgical hypotension feature information, wherein the historical surgical hypotension feature information is used to characterize the dynamic change rate of the user's physiological indicators, surgical operation factors, and anesthesia management factors; Processing the target user's real-time surgical preset information to generate the target user's real-time physiological characteristic parameters during the operation; Generate surgical hypotension prediction information based on historical surgical hypotension characteristic information and real-time physiological characteristic parameters of the target user during surgery; The surgical hypotension prediction information is processed to generate interval prediction subtasks and intraoperative prediction subtasks.

3. The method according to claim 2, wherein The interval prediction subtask and the intraoperative prediction subtask are processed to generate a preset prediction feature vector, including; The interval prediction subtask and the intraoperative prediction subtask are processed to generate target feature information, where the target feature information includes the real-time inter-operative time interval value, the target user's physiological indicator mean, the physiological indicator change slope, the intraoperative operation time, and the anesthesia depth fluctuation value; Processing the target feature information to generate a first feature association value and a second feature association value, wherein the first feature association value is used to characterize the correlation between the surgical interval duration pattern and the change in the physiological characteristic, and the second feature association value is used to characterize the correlation between the surgical characteristic and the risk of hypotension; The first feature association value and the second feature association are processed to generate a preset prediction feature vector.

4. The method according to claim 3, wherein The target user's physiological motion parameter information is processed to generate the target user's blood pressure limit factor, including: Process the target user's physiological motion parameter information to generate dynamic respiratory waveform features, dynamic electrocardiogram signal features, dynamic blood oxygen saturation features, and dynamic end-tidal carbon dioxide partial pressure features; Processing dynamic respiratory waveform features, dynamic electrocardiogram signal features, dynamic blood oxygen saturation features, and dynamic end-tidal carbon dioxide partial pressure features to generate characteristic parameter information for different respiratory cycles; The characteristic parameter information of different respiratory cycles is processed to generate the blood pressure limit factor of the target user.

5. The method according to claim 1, wherein The physiological status information of the target user is processed to generate the dynamic blood pressure influencing factors of the target user, including: Processing the target user's physiological state information to generate waveform parameter characteristic information, respiratory parameter characteristic information, and cardiac motion parameter information; The waveform parameter characteristic information is processed to generate the coefficient of variation of systolic blood pressure, coefficient of variation of diastolic blood pressure, coefficient of variation of mean arterial pressure slope, low-frequency to high-frequency power ratio of heart rate variability, integral of ST segment depression or elevation amplitude, coefficient of variation of QRS complex duration, coefficient of variation of pulse wave transmission time, coefficient of variation of perfusion index, coefficient of variation of end-tidal carbon dioxide partial pressure, coefficient of variation of carbon dioxide waveform inspiratory phase slope, and coefficient of variation of expiratory phase slope; Process the characteristic information of respiratory parameters to generate the standard deviation of respiratory frequency, coefficient of variation of respiratory depth, and coefficient of variation of the ratio of inspiratory time to expiratory time; Process cardiac motion parameter information to generate heart rate standard deviation, arrhythmia score, and myocardial contractility index; Processing waveform parameter characteristic information, respiratory parameter characteristic information, and cardiac motion parameter information to generate a dynamic blood pressure impact value; The dynamic blood pressure impact value is processed to generate the dynamic blood pressure impact factor of the target user.

6. The method according to claim 1, wherein Based on the target hypotension warning model, the preset prediction feature vector, the target user's blood pressure limit factor, and the target user's dynamic blood pressure influencing factor are processed to generate hypotension warning information for the target user, including: Based on the target hypotension warning model, the preset prediction feature vector, the target user's blood pressure limit factor and the target user's dynamic blood pressure influencing factor are subjected to feature fusion processing to generate an initial feature matrix; Perform dimensionality reduction on the initial feature matrix to generate the initial probability value of low blood pressure for the target user; The initial probability value of hypotension of the target user is processed based on the real-time surgery preset information of the target user to generate hypotension information of the target user within a preset time period.

7. A warning device for intraoperative hypotension, characterized in that: The device includes: An acquisition module is used to acquire the target user's surgical information, target user's physiological motion parameter information, and target user's physiological state information within a preset time period; a processing module configured to process the target user's surgical information within a preset time period to generate an interval prediction subtask and an intraoperative prediction subtask; process the interval prediction subtask and the intraoperative prediction subtask to generate a preset prediction feature vector; process the target user's physiological motion parameter information to generate the target user's blood pressure limiting factor; and process the target user's physiological state information to generate the target user's dynamic blood pressure influencing factor; Based on the target hypotension warning model, the preset prediction feature vector, the target user's blood pressure limit factor and the target user's dynamic blood pressure influencing factor are processed to generate hypotension warning information for the target user.

8. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions for the first processor; The first processor is configured to execute the early warning method for intraoperative hypotension according to any one of claims 1 to 6 by executing executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the method for early warning of intraoperative hypotension according to any one of claims 1 to 6 is implemented.

Images