Surgical anesthesia intraoperative risk early warning method based on ai multi-modal data fusion

Abstract

The present application relates to the technical field of clinical risk early warning, in particular to a surgery anesthesia intraoperative risk early warning method based on AI multi-modal data fusion, comprising the following steps: constructing an ideal steady-state residence model based on anesthesia records, monitoring signs in combination with blood loss and calculating cumulative load of drug metabolism, evaluating physiological steady-state confidence, comparing benchmarks to determine implicit compensation failure, and generating compensation depletion risk early warning instructions. In the present application, a probability density model representing the physiological steady-state maintenance capability is constructed, the risk assessment dimension is expanded from a single numerical amplitude to a time residence length, the blood loss and drug load data accumulated in real time during the operation are used to perform dynamic time domain compression and coordinate translation transformation on the benchmark distribution curve, the implicit failure risk is identified in advance when the vital sign readings have not yet exceeded the normal range but the compensation potential is about to be depleted, the steady-state collapse trend caused by the time cumulative effect is effectively quantified, and the forward-looking of risk identification is improved.

Description

Technical Field

[0001] This invention relates to the field of clinical risk warning technology, and in particular to a surgical anesthesia risk warning method based on AI multimodal data fusion. Background Technology

[0002] The field of clinical risk warning technology encompasses various technical means and systems that identify potential medical safety hazards and issue warnings to medical staff through real-time monitoring and analysis of patient vital signs, laboratory test results, and clinical diagnosis and treatment process data. This field is dedicated to building a comprehensive monitoring system that integrates multi-dimensional information such as patient medical records, current physiological parameters, and drug responses. It utilizes thresholds set by statistical rules or clinical guidelines to determine the trend of changes in the patient's condition, assisting doctors in making rapid decisions and taking intervention measures during the perioperative period, intensive care, and general ward nursing. Traditional surgical anesthesia risk warning methods refer to the real-time monitoring and abnormal alerting process for fluctuations in the patient's physiological state during the anesthesia induction, maintenance, and recovery periods. Existing technologies rely on anesthesiologists directly observing the electrocardiogram waveforms, invasive or non-invasive blood pressure readings, blood oxygen saturation percentage, and end-tidal carbon dioxide partial pressure data displayed on the monitor screen. These single-dimensional physiological indicators are compared with pre-set fixed upper and lower limits on the anesthesia machine or monitoring equipment. When the real-time reading of a certain indicator exceeds the set range, a Boolean logic judgment mechanism is triggered, thereby driving an audible and visual alarm to emit a warning sound and flashing signal.

[0003] Existing technologies rely on mechanical comparisons of physiological indicators at a single moment with fixed thresholds. This transient monitoring mode severs the continuous evolution of vital signs over time, ignores the implicit depletion of the body's physiological compensatory capacity during prolonged surgery, and the fixed threshold judgment logic is unable to perceive the critical state where the values ​​are within the normal range but the body's ability to cope with cumulative load is approaching its limit. This results in a lack of sensitivity to cumulative damage caused by continuous micro-blood loss or drug accumulation, causing risk warnings to occur after the complete imbalance of physiological homeostasis, missing the opportunity for early intervention, and exhibiting obvious blind spots and response delays. Summary of the Invention

[0004] To address the technical problems existing in the prior art, embodiments of the present invention provide a method for intraoperative risk warning in surgical anesthesia based on AI multimodal data fusion, comprising the following steps: To achieve the above objectives, the present invention adopts the following technical solution: a surgical anesthesia risk early warning method based on AI multimodal data fusion, comprising the following steps: S1: Based on the patient's anesthesia medical records, identify continuous stable periods in which arterial pressure and bispectral index of electroencephalogram maintain the standard range, perform statistical analysis on the distribution frequency of stable periods in differentiated time intervals, connect the frequency points to form a probability density shape, and construct an ideal steady-state residence time distribution model; S2: Referring to the normal parameter range defined by the ideal steady-state residence time distribution model, monitor real-time vital signs, track the real-time steady-state duration, and simultaneously collect intraoperative blood loss data and anesthetic drug infusion data. Perform cumulative calculation on intraoperative blood loss and integrate drug concentration at the pharmacokinetic level to generate quantitative indicators of intraoperative cumulative physiological load. S3: Call the intraoperative cumulative physiological load quantification index, calculate the compression ratio coefficient of the distribution area on the time axis based on the load intensity, perform coordinate transformation operation, and obtain the dynamic drift survival probability density curve; S4: Using the dynamic drift survival probability density curve, locate the coordinate point corresponding to the duration of real-time steady state, calculate the coverage area of ​​the time region after the coordinate point, assess the probability that the body will continue to maintain a steady state, and obtain the confidence index of physiological homeostasis maintenance.

[0005] As a further aspect of the present invention, the ideal steady-state residence time distribution model includes a baseline probability density function parameter, a confidence boundary of the steady-state time interval, and a baseline value of the integral area. The quantitative index of intraoperative cumulative physiological load includes a blood loss load weighted component, a drug metabolism residue integral component, and a real-time physiological stress score. The dynamic drift survival probability density curve includes a time axis coordinate set after compression transformation, a corrected instantaneous survival probability value, and a dynamic decay trend slope. The physiological homeostasis maintenance confidence index includes a residual steady-state probability value under real-time conditions, a normalized percentage of compensatory capacity margin, and a homeostasis maintenance failure risk rate.

[0006] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Based on the patient's anesthesia medical records, extract the entire arterial pressure monitoring numerical sequence and EEG bispectral index waveform data, call the preset physiological parameter standard maintenance range, compare the overlap between the arterial pressure monitoring numerical sequence and the physiological parameter standard maintenance range one by one, locate the start time stamp and end time stamp when the values ​​of both are maintained within the physiological parameter standard maintenance range, extract the continuous time span between the two time stamps, and obtain the dual-parameter synchronous steady-state time domain segment. S102: Call the dual-parameter synchronous steady-state time domain segment, calculate the segment's time length value, establish a differentiated time interval histogram statistical box, map the time length value to the corresponding histogram statistical box, accumulate the distribution frequency, use the time interval center value as the horizontal axis and the distribution frequency as the vertical axis to draw discrete points, connect adjacent discrete frequency points to form a continuous morphological trajectory, and generate a steady-state duration frequency envelope curve; S103: For the steady-state duration frequency envelope curve, define the closed region enclosed by the curve edge trajectory and the time coordinate axis, perform area calculation on the closed region, obtain the quantitative value characterizing the body's baseline tolerance, use the quantitative value to normalize and correct the amplitude of the curve's vertical axis, determine the probability density function mapping relationship under differentiated durations, and construct an ideal steady-state residence time distribution model.

[0007] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Call the normal parameter range defined by the ideal steady-state residence time distribution model, read the patient's arterial pressure and EEG bifrequency index in real time through the monitoring interface, compare the real-time values ​​with the normal parameter range, determine the steady state and record the duration, generate the real-time steady-state duration, and simultaneously collect the suction bottle fluid volume scale and the surgical gauze weight gain value. Perform cumulative summation on the blood loss data collected in a single session according to the time series to obtain the total cumulative blood loss during the operation. S202: Based on the total cumulative blood loss during the operation, monitor the real-time flow rate setting and drug addition dose of the anesthetic pump, set the drug metabolism attenuation coefficient and compartment distribution volume parameters, perform convolution processing and plasma concentration simulation on the pump data, calculate the residual active drug concentration in the real-time effect compartment, and perform integral integration operation on the cumulative effect of the active concentration over time to obtain integrated pharmacokinetic drug concentration data. S203: Call the integrated pharmacokinetic drug concentration data, set a first weighting coefficient based on the effect of blood loss factors on the loss of circulating volume, set a second weighting coefficient based on the inhibitory intensity of drug factors on the central nervous system, perform weighted multiplication operations on the two data, and perform linear superposition and summation operations on the weighted blood loss value and the weighted integrated drug value to generate a quantitative index of intraoperative cumulative physiological load.

[0008] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Call the intraoperative cumulative physiological load quantification index, retrieve the mapping association list between load intensity and time axis scaling ratio, lock the corresponding time compression ratio coefficient according to the load index value, and simultaneously set the step displacement value of the distribution area moving towards the coordinate origin according to the load magnitude. Perform vectorization combination operation on the compression coefficient and step displacement value to generate time axis compression and translation transformation parameter vector. S302: Call the time axis compression and translation transformation parameter vector, extract the horizontal axis coordinate point set that defines the steady state duration in the ideal steady state residence time distribution model, perform multiplicative shrinkage operation on the horizontal axis coordinate point set using the compression ratio coefficient, and reduce the distribution span of the steady state duration through coordinate transformation operation while keeping the vertical axis probability density amplitude constant, reconstruct the correspondence between the shrunken coordinate point set and the probability density, and generate a time-domain shrinkage distribution morphology dataset; S303: For the time-domain contraction distribution pattern dataset, read the step displacement values ​​in the time axis compression and translation transformation parameter vector, perform subtraction operation on the time coordinate points, implement the overall translation operation towards the coordinate zero point, identify the coordinate points that fall into the negative value interval after translation and perform zero-truncation processing, redraw the distribution envelope edge trajectory according to the updated coordinate point sequence, and obtain the dynamic drift survival probability density curve.

[0009] As a further aspect of the present invention, the mapping association list of load intensity and time axis scaling ratio adopts a discrete interval table structure that is sorted in ascending order of load index value intervals. Each load index value interval corresponds to a unique compression ratio coefficient, and the compression ratio coefficient is limited to a positive number less than or equal to one. The step displacement value of the distribution area moving towards the origin of the coordinate system is determined by matching the load magnitude with a preset step displacement scale, and the step displacement value is a non-negative real number. Specifically, the vectorization combination operation involves encapsulating the compression ratio coefficient and the step displacement value into the same parameter vector in a fixed-dimensional order. The zero-truncation process is defined as directly assigning zero to time coordinate points that are less than zero after subtraction, keeping the original values ​​of the remaining time coordinate points unchanged, and redrawing the distribution envelope edge trajectory based on the one-to-one correspondence between the time coordinate point sequence after zero-truncation and the original probability density value.

[0010] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Call the dynamic drift survival probability density curve, map the duration value as the horizontal axis index to the time axis of the curve, anchor the position of the real-time node in the probability distribution, delineate the lag time region from the real-time node to the zero point of the probability density of the curve, perform definite integral operation on the range enclosed by the probability density envelope trajectory and the time axis within the lag time region, and generate the remaining steady-state survival coverage area. S402: For the remaining steady-state coverage area, traverse the complete span of the time axis from zero to the convergence point of the curve edge, extract the probability density amplitude corresponding to the discrete time points within the span, calculate the total amount of the closed geometric region enclosed by the curve edge trajectory and the time coordinate axis, quantify the distribution weight of the probability density throughout the entire life cycle, and generate the total area of ​​the global probability distribution. S403: Call the total area of ​​the global probability distribution to construct the conditional probability evaluation logic. Perform a division operation with the remaining area value as the numerator and the total area value as the denominator to calculate the probability density ratio after the real-time node, quantify the probability that the organism will continue to maintain a stable state under real-time load conditions, and generate the physiological homeostasis maintenance confidence index.

[0011] As a further aspect of the present invention, the duration value is limited to a non-negative time scalar calculated from the zero point of the time axis of the dynamic drift survival probability density curve, and the position of the real-time node in the probability distribution is determined by the one-to-one correspondence between the duration value and the horizontal axis of the dynamic drift survival probability density curve. The lag time region is defined as a closed interval between the horizontal coordinate of the real-time node and the intersection of the dynamic drift survival probability density curve and the time axis, and the probability density zero point is the time coordinate position where the probability density amplitude first equals zero. The definite integral operation is implemented by numerical integration, which accumulates and sums the probability density amplitudes corresponding to discrete time coordinate points within the lag time region. The convergence point of the curve edge is defined as the time coordinate position where the probability density amplitude is continuously lower than a preset threshold and remains unchanged, and the range of the physiological homeostasis maintenance confidence index is limited to a closed interval from zero to one.

[0012] As a further aspect of the present invention, the method further includes step S5: S5: Compare the physiological homeostasis maintenance confidence index with the preset safety warning benchmark, identify the target situation where the physiological homeostasis maintenance confidence index is lower than the safety warning benchmark, determine that the patient is in the latent compensatory failure stage, transmit a signal including the intervention time window to the interactive terminal, and generate a compensatory depletion risk warning instruction. The risk warning instruction for the depletion of compensation includes the warning level of implicit compensation failure, the start and end points of the time window for intervention, and the classification description code of the risk status.

[0013] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Call the physiological homeostasis maintenance confidence index, read the preset safety warning benchmark value, perform a numerical comparison operation between the physiological homeostasis maintenance confidence index and the safety warning benchmark, monitor the difference and polarity characteristics between the two in real time, determine the target abnormal situation where the physiological homeostasis maintenance confidence index is lower than the safety warning benchmark, identify the real-time physiological compensation level of the body, mark the physiological compensation level attribute as a latent failure state, and generate a latent compensation failure state judgment identifier; S502: For the implicit compensatory failure state determination identifier, the tail convergence feature of the dynamic drift survival probability density curve is retrieved again, the remaining time span before the probability density completely returns to zero is calculated, the time span is defined as the effective buffer period for clinical operation, the time value of the buffer period is binary encoded and formatted and encapsulated to obtain emergency intervention time window signal data. S503: Call the emergency intervention time window signal data, establish a real-time data transmission channel with the external interactive terminal, map the emergency intervention time window signal to the preset alarm protocol template field, synchronously integrate the risk level code and sound and light drive control parameters, perform instruction-based packaging and integrity verification on the alarm elements, and generate a compensation depletion risk warning instruction.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a probability density model characterizing the ability to maintain physiological homeostasis is constructed, expanding the risk assessment dimension from a single numerical amplitude to the duration of time residence. By using intraoperatively accumulated blood loss and drug load data, dynamic time-domain compression and coordinate translation transformation are performed on the baseline distribution curve to accurately map the contraction trend of the body's compensatory boundary under high load conditions. Based on the corrected survival probability, the confidence level of maintaining physiological homeostasis is calculated, and the risk of latent failure is identified in advance when the vital signs reading has not exceeded the normal range but the compensatory potential is about to be exhausted. This effectively quantifies the homeostasis collapse trend caused by the cumulative effect of time and improves the foresight of risk identification. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0019] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0020] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Please see Figure 1 This invention provides a method for intraoperative risk warning in surgical anesthesia based on AI multimodal data fusion, comprising the following steps: S1: Based on the patient's anesthesia medical records, identify continuous stable periods in which arterial pressure and bispectral index of electroencephalogram maintain the standard range, perform statistical analysis on the distribution frequency of stable periods in differentiated time intervals, connect frequency points to form a probability density shape, perform area integration on the area covered by the probability density shape, quantify the baseline tolerance capacity, and construct an ideal steady-state residence time distribution model. S2: Based on the normal parameter range defined by the ideal steady-state residence time distribution model, monitor real-time vital signs, track the duration of real-time steady state, and simultaneously collect intraoperative blood loss data and anesthetic drug infusion data. Perform cumulative calculation on intraoperative blood loss and integrate drug concentration at the pharmacokinetic level. Sum the weighted intraoperative blood loss data and the weighted integrated drug data to generate a quantitative index of intraoperative cumulative physiological load. S3: Call the intraoperative cumulative physiological load quantification index, calculate the compression ratio coefficient of the distribution area on the time axis based on the load intensity, perform coordinate transformation operation, shrink the time scale of the ideal steady-state residence time distribution model, and shift the overall distribution area towards the zero point of the time axis according to the load size to obtain the dynamic drift survival probability density curve. S4: Using the dynamic drift survival probability density curve, locate the coordinate point corresponding to the duration of real-time steady state, use the numerical integration method to calculate the coverage area of ​​the time region after the coordinate point, divide the coverage area by the total area of ​​the distribution area, assess the probability that the organism will continue to maintain a steady state, and obtain the confidence index of physiological homeostasis maintenance. S5: Compare the confidence index of maintaining physiological homeostasis with the preset safety threshold, identify the target situation where the confidence index of maintaining physiological homeostasis is lower than the safety threshold, determine that the patient is in the latent compensatory failure stage, transmit signals including the intervention time window to the interactive terminal, and generate a risk warning instruction for the depletion of compensation. The ideal steady-state residence time distribution model includes the baseline probability density function parameters, the confidence boundary of the steady-state time interval, and the baseline value of the integral area. The quantitative indicators of intraoperative cumulative physiological load include the weighted component of blood loss load, the integral component of drug metabolism residue, and the real-time physiological stress score. The dynamic drift survival probability density curve includes the time axis coordinate set after compression transformation, the corrected instantaneous survival probability value, and the slope of the dynamic decay trend. The confidence index of physiological homeostasis maintenance includes the remaining steady-state probability value under real-time conditions, the normalized percentage of compensatory capacity surplus, and the risk rate of homeostasis maintenance failure. The early warning instructions for the risk of compensation depletion include the warning level of implicit compensation failure, the start and end points of the time window for intervention, and the classification description code of the risk status.

[0023] Please see Figure 2 The specific steps of S1 are as follows: S101: Based on the patient's anesthesia medical records, extract the entire arterial pressure monitoring numerical sequence and EEG bispectral index waveform data, call the preset physiological parameter standard maintenance range, compare the overlap between the arterial pressure monitoring numerical sequence and the physiological parameter standard maintenance range one by one, locate the start time stamp and end time stamp when the values ​​of both are maintained within the physiological parameter standard maintenance range, extract the continuous time span between the two time stamps, and obtain the dual-parameter synchronous steady-state time domain segment. A connection is established between the hospital information system and the anesthesia monitor via their communication interface. Utilizing the HL7 medical data exchange protocol, the raw data stream during general anesthesia is captured at a sampling frequency of 1 Hz. For the acquired arterial pressure data, a Kalman filter algorithm is used to remove high-frequency noise interference caused by changes in body position or sensor vibration, retaining low-frequency physiological trend signals. For the bispectral index data from electroencephalography (EEG), electromyographic interference artifacts are identified and eliminated to ensure data purity. Pre-set physiological parameter standards are then invoked, based on guidelines from the American College of Anesthesiologists (ACS), for example, the standard maintenance range for mean arterial pressure is set to 65. The range was from 105 mmHg to 105 mmHg, with the standard maintenance range of the bispectral index (BSE) set to 40 to 60. A point-by-point scanning comparison was performed, aligning the cleaned arterial pressure sequence with the BSE sequence on the same time axis. It was determined whether the value at each moment fell within the standard maintenance range. When both parameters were detected to enter the standard range at the same time, that moment was marked as the start timestamp; when either parameter deviated from the standard range, that moment was marked as the end timestamp. The continuous data segment between the start and end timestamps was extracted to obtain the synchronous steady-state time domain segment of the two parameters.

[0024] S102: Call the two-parameter synchronous steady-state time domain segment, calculate the segment's time length value, establish a differentiated time interval histogram statistical bin, map the time length value to the corresponding histogram statistical bin, accumulate the distribution frequency, use the time interval center value as the horizontal axis and the distribution frequency as the vertical axis to draw discrete points, connect adjacent discrete frequency points to form a continuous morphological trajectory, and generate a steady-state duration frequency envelope curve; The duration of each segment is calculated as the difference between the end timestamp and the start timestamp. If a segment starts 30 minutes after the start of surgery and ends 75 minutes after the start of surgery, the duration of that segment is 45 minutes. A differential time interval histogram statistical bin is established, with the bin interval set to 5 minutes, covering the range from 0 minutes to the maximum surgery duration. The extracted steady-state segment durations are traversed, and their values ​​are mapped to the corresponding histogram statistical bins. For example, the 45-minute segment mentioned above will be included in the "40 minutes to 45 minutes" statistical bin. The cumulative distribution frequency is statistically analyzed for the sample size in each bin. The center value of each time interval (e.g., 42.5 minutes) is used as the x-axis, and the corresponding distribution frequency is used as the y-axis. Discrete points are plotted in a two-dimensional coordinate system. A cubic spline interpolation algorithm is used to connect adjacent discrete frequency points to eliminate data jitter and form a smooth and continuous morphological trajectory, generating a steady-state duration frequency envelope curve.

[0025] S103: For the steady-state duration frequency envelope curve, define the closed region enclosed by the curve edge trajectory and the time coordinate axis, perform area calculation on the closed region, obtain the quantitative value characterizing the body's baseline tolerance, use the quantitative value to normalize and correct the amplitude of the curve's vertical axis, determine the probability density function mapping relationship under differentiated durations, and construct an ideal steady-state residence time distribution model. The closed region enclosed by the curve's edge trajectory and the time axis (horizontal axis) is defined. The trapezoidal numerical integration method is used to perform area calculations on this closed region. Specifically, the time axis is divided into tiny differential units, and the area of ​​the rectangle corresponding to each unit is calculated and summed to obtain a quantitative value characterizing the body's baseline tolerance. This value reflects the patient's overall potential to maintain physiological homeostasis under ideal conditions. This quantitative value is used to normalize the amplitude of the curve's vertical axis. That is, the vertical coordinate value of each frequency point on the curve is divided by this quantitative value, making the total area under the curve equal to 1. After this operation, the vertical coordinate is converted from absolute frequency to probability density, determining the probability density function mapping relationship under differentiated durations. The corrected curve data is stored as a standardized dataset. If the calculated total area of ​​the closed region is 500 and the original frequency of a certain point is 50, the normalized probability density value is 0.1, intuitively describing the inherent probability distribution of the patient's ability to maintain a specific duration of homeostasis without additional load impact, thus constructing an ideal homeostasis residence time distribution model.

[0026] Please see Figure 3 The specific steps of S2 are as follows: S201: Call the normal parameter range defined by the ideal steady-state residence time distribution model, read the patient's arterial pressure and EEG bispectral index in real time through the monitoring interface, compare the real-time values ​​with the normal parameter range, determine the steady state and record the duration, generate the real-time steady-state duration, and simultaneously collect the suction bottle fluid volume scale and the surgical gauze weight gain value. Perform cumulative summation on the blood loss data collected in a single session according to the time series to obtain the total cumulative blood loss during the operation. Based on the defined normal parameter range (i.e., arterial pressure 65 mmHg to 105 mmHg and bispectral index 40 to 60), the patient's current arterial pressure and bispectral index are read in real time at a refresh rate of 500 milliseconds through the monitoring interface. The real-time values ​​are compared with the normal parameter range. If both are within the range, a timer is started to record the duration and generate the real-time steady-state duration. At the same time, the scale change of the surgical aspiration bottle volume and the weight gain of the surgical gauze are collected synchronously through a high-precision electronic scale sensor. The aspiration bottle volume data needs to be reduced by the known amount of intraoperative irrigation fluid. The gauze weight gain is obtained by measuring the weight of the gauze after use and subtracting the standard weight of the dry gauze before use. At a time series interval of 1 minute, the blood loss data (including liquid blood loss and gauze seepage) collected in a single session are summed. The gauze weight gain is converted into volume according to the blood density (1.06 g / mL) and added to the pure blood volume in the aspiration bottle to obtain the total cumulative blood loss during the operation.

[0027] S202: Based on the total cumulative blood loss during the operation, monitor the real-time flow rate setting and drug addition dose of the anesthetic pump, set the drug metabolism attenuation coefficient and compartment distribution volume parameters, perform convolution processing and plasma concentration simulation on the pump data, calculate the residual active drug concentration in the real-time effect compartment, and perform integral integration operation on the cumulative effect of the active concentration over time to obtain integrated pharmacokinetic drug concentration data. The real-time flow rate settings (unit: ml / h) and drug supplementation dose (unit: mg) of the anesthetic infusion pump were monitored. The drug metabolism decay coefficient and compartment distribution volume parameters were set. Taking propofol as an example, the central compartment distribution volume was set to 228 ml / kg and the elimination rate constant was set to 0.119 m / min. The pharmacokinetic three-compartment model algorithm was used to perform convolution processing on the infusion data. The drug infusion rate sequence was used as the input function and convolved with the unit impulse response function corrected based on the patient's weight and age to simulate the change of blood drug concentration over time. The plasma-effect compartment equilibrium rate constant was introduced to calculate the residual active drug concentration in the real-time effect compartment. The cumulative effect of the active concentration over time was integrated, that is, the area under the concentration-time curve from the start of anesthesia induction to the current moment was calculated to obtain the integrated pharmacokinetic drug concentration data.

[0028] S203: Call the integrated pharmacokinetic drug concentration data, set the first weighting coefficient based on the effect of blood loss factors on the loss of circulating volume, set the second weighting coefficient based on the inhibitory intensity of drug factors on the central nervous system, perform weighted multiplication operations on the two data, and perform linear superposition and summation operation on the weighted blood loss value and the weighted integrated drug value to generate the quantitative index of intraoperative cumulative physiological load. Based on regression analysis of batch clinical retrospective data, a first weighting coefficient of 0.002 (corresponding to the weight of the disruption to steady state per milliliter of blood loss) and a second weighting coefficient of 0.05 (corresponding to the weight of the suppression of steady state per unit of drug concentration integral) were set. Weighted multiplication operations were performed on the two data points: the weighted blood loss value equals the total intraoperative blood loss multiplied by the first weighting coefficient; the weighted drug integration value equals the pharmacokinetic drug concentration integration data multiplied by the second weighting coefficient. The weighted blood loss value and the weighted drug integration value were then linearly summed to generate a quantitative index of intraoperative cumulative physiological load. Table 1 shows an example of intermediate data collected and calculated at a certain moment: Table 1: Parameters for Calculating Intraoperative Physiological Load Parameter name numerical values unit Remark Total cumulative blood loss during surgery 450 milliliters Suction bottle + gauze statistics First weighting coefficient 0.002 Dimensionless Loop Capacity Weight Integrated pharmacokinetic drug concentration data 1200 Nak·h / ml Concentration-time integral Second weighting coefficient 0.05 Dimensionless Neural Inhibition Weights Referring to the data in Table 1, the following calculations were performed: the weighted blood loss value was 450 multiplied by 0.002, which equaled 0.9; the weighted drug integration value was 1200 multiplied by 0.05, which equaled 60. The final intraoperative cumulative physiological load quantification index was 0.9 plus 60, which equaled 60.9. The higher the value of this index, the heavier the internal and external load on the body, and the greater the difficulty in maintaining homeostasis. The advantage of this calculation logic is that it transforms physical quantities of different dimensions (blood loss volume and drug concentration) into a unified dimensionless load index, thereby realizing the comprehensive quantification of multi-source physiological pressure.

[0029] Please see Figure 4 The specific steps of S3 are as follows: S301: Call the intraoperative cumulative physiological load quantification index, retrieve the mapping association list between load intensity and time axis scaling ratio, lock the corresponding time compression ratio coefficient according to the load index value, and simultaneously set the step displacement value of the distribution area moving towards the coordinate origin according to the load magnitude. Perform vectorized combination operation on the compression coefficient and step displacement value to generate time axis compression and translation transformation parameter vector. The list is based on large-sample physiological limit stress test data. It defines a non-linear relationship between the load index and the degree of time compression. Based on the load index value calculated above (e.g., 60.9), the corresponding time compression ratio coefficient and step size displacement value are locked. The rules are set as follows: for every 10 units increase in the load index, the compression ratio coefficient decreases by 0.05 from the baseline value of 1.0, and the step size displacement value increases by 2 minutes. For example, when the load index is 60.9, the compression ratio coefficient is determined to be 0.695 and the step size displacement value is 12.18 minutes through linear interpolation. These two values ​​are then vectorized and combined to generate a time axis compression and translation transformation parameter vector containing scaling and translation factors.

[0030] S302: Call the time axis compression and translation transformation parameter vector, extract the horizontal axis coordinate point set that defines the steady state duration in the ideal steady state residence time distribution model, perform multiplicative shrinkage operation on the horizontal axis coordinate point set using the compression ratio coefficient, and reduce the distribution span of the steady state duration through coordinate transformation operation while keeping the vertical axis probability density amplitude constant, reconstruct the correspondence between the shrunken coordinate point set and the probability density, and generate a time-domain shrunken distribution morphology dataset; The point set consists of a series of discrete values ​​representing time (such as 10 minutes, 11 minutes, 12 minutes, etc.). A multiplicative shrinkage operation is performed on the horizontal axis coordinate point set using a compression scaling factor. Specifically, each time coordinate value in the point set is multiplied by the compression scaling factor (0.695). The physical meaning of this operation is that under high load conditions, the probability of the organism maintaining the same steady state duration is compressed, and the long-term steady state that was originally highly probable becomes difficult to achieve. Under the benchmark of keeping the vertical axis probability density amplitude constant, the distribution span of the steady state duration is reduced through coordinate transformation operations, and the correspondence between the shrunken coordinate point set and the probability density is reconstructed to generate a time-domain shrunken distribution morphology dataset.

[0031] S303: For the time-domain shrinkage distribution pattern dataset, read the step displacement values ​​in the time axis compression and translation transformation parameter vector, perform subtraction operation on the time coordinate points, implement the overall translation operation towards the coordinate zero point, identify the coordinate points that fall into the negative value interval after translation and perform zero-truncation processing, redraw the distribution envelope edge trajectory according to the updated coordinate point sequence, and obtain the dynamic drift survival probability density curve. Read the step displacement value (12.18 minutes) in the time axis compression and translation transformation parameter vector, perform subtraction operation on the condensed time coordinate points, that is, subtract 12.18 from each new coordinate value, and perform a global translation operation towards the coordinate zero point to simulate the early arrival of the steady-state collapse critical point caused by load accumulation. Identify the coordinate points that fall into the negative value range after translation. If the calculation result of a coordinate point is less than 0, perform zero-truncation processing and force it to be set to 0. According to the updated coordinate point sequence, redraw the distribution envelope edge trajectory. At this time, the original normal or skewed distribution curve will be "squeezed" and "translated" to the left to form a curve with a steeper shape and a shorter time span, and obtain the dynamic drift survival probability density curve. Table 2 shows an example of data comparison before and after coordinate transformation: Table 2: Steady-state time-time coordinate transformation comparison table Original time coordinates (minutes) Compression ratio coefficient Mid-contraction coordinates (minutes) Step size displacement (minutes) Final coordinate transformation (minutes) 60 0.695 41.7 12.18 29.52 100 0.695 69.5 12.18 57.32 15 0.695 10.425 12.18 0 (Truncation) Table 2 presents the test results of the embodiment. The experimental results show that under a physiological load of 60.9 units, the steady-state consumption level that originally required 60 minutes to reach can now be reached in only 29.52 minutes, revealing the significant weakening effect of load on the ability to maintain steady state.

[0032] Please see Figure 5 The specific steps of S4 are as follows: S401: Call the dynamic drift survival probability density curve, map the duration value as the horizontal axis index to the time axis of the curve, anchor the position of the real-time node in the probability distribution, delineate the lag time region from the real-time node to the zero point of the curve probability density, perform definite integral operation on the range enclosed by the probability density envelope trajectory and the time axis within the lag time region, and generate the remaining steady-state survival coverage area. The real-time steady-state duration is used as the horizontal axis index and mapped to the time axis of the curve. If the patient has been stable for 25 minutes, the 25-minute mark is anchored on the time axis of the dynamic curve to determine the position of the real-time node in the probability distribution. The lag time region from the real-time node (25 minutes) to the zero point of the probability density of the curve (i.e., the intersection of the right side of the curve and the horizontal axis, assumed to be 80 minutes) is defined. A definite integral operation is performed on the range enclosed by the probability density envelope trajectory and the time axis within the lag time region. Specifically, the area under the probability density function curve in the interval from 25 minutes to 80 minutes is calculated. This area represents the cumulative probability mass that the body can continue to maintain steady state after the current moment, generating the remaining steady-state coverage area.

[0033] S402: For the remaining steady-state coverage area, traverse the complete span of the time axis from zero to the convergence point of the curve edge, extract the probability density amplitude corresponding to the discrete time points within the span, calculate the total amount of the closed geometric region enclosed by the curve edge trajectory and the time coordinate axis, quantify the distribution weight of the probability density throughout the entire life cycle, and generate the total area of ​​the global probability distribution. Traverse the entire span of the time axis from zero to the convergence point of the curve edge (80 minutes), extract the probability density amplitude corresponding to the discrete time points within the span, calculate the total amount of the closed geometric region enclosed by the curve edge trajectory and the time coordinate axis, that is, integrate the probability density function over the entire domain from 0 to 80 minutes. Since the previous steps have been normalized, but the total area is deformed after compression and translation transformation, it is necessary to recalculate to quantify the distribution weight of the probability density over the entire life cycle and generate the total area of ​​the global probability distribution.

[0034] S403: Call the total area of ​​the global probability distribution, construct the conditional probability evaluation logic, use the remaining area value as the numerator and the total area value of the global region as the denominator to perform a division operation, calculate the probability density ratio after the real-time node, quantify the probability that the organism will continue to maintain a stable state under real-time load conditions, and generate the physiological homeostasis maintenance confidence index. A conditional probability assessment logic is constructed to answer the question, "Given that an organism has maintained a steady state for time t, what is the probability that it can continue to maintain it?" The remaining area value is used as the numerator, and the total area value of the entire region is used as the denominator. A division operation is performed. If the calculated remaining steady-state coverage area is 0.35 and the total area of ​​the probability distribution of the entire region is 0.95, then the operation of 0.35 divided by 0.95 is performed, and the result is approximately equal to 0.368. This calculation result is used to quantify the probability that the organism will continue to maintain a steady state under real-time load conditions, and a physiological homeostasis maintenance confidence index is generated.

[0035] Please see Figure 6 The specific steps of S5 are as follows: S501: Call the physiological homeostasis maintenance confidence index, read the preset safety warning benchmark value, compare the physiological homeostasis maintenance confidence index with the safety warning benchmark value, monitor the difference and polarity characteristics between the two in real time, determine the target abnormal situation where the physiological homeostasis maintenance confidence index is lower than the safety warning benchmark, identify the body's real-time physiological compensation level, mark the physiological compensation level attribute as a latent failure state, and generate a latent compensation failure state judgment label; The preset safety warning baseline value is read. This baseline value is set according to the anesthesia safety operation specifications. For example, it is set to 0.4, which means that when the conditional probability of maintaining homeostasis is less than 40%, the body's compensatory ability is considered to be nearly exhausted. The physiological homeostasis maintenance confidence index (e.g., 0.368 calculated above) is compared with the safety warning baseline (0.4). The difference and polarity characteristics between the two are monitored in real time. The difference is calculated as 0.368 minus 0.4 equals -0.032. Since the difference is negative, the target abnormal situation of the physiological homeostasis maintenance confidence index being lower than the safety warning baseline is identified. The real-time physiological compensation level of the body is identified as insufficient to resist the current physiological load. The physiological compensation level attribute is marked as a latent failure state, and a latent compensation failure state judgment label is generated.

[0036] S502: For the latent compensatory failure status identification marker, the tail convergence characteristics of the dynamic drift survival probability density curve are retrieved again, the remaining time span before the probability density completely returns to zero is calculated, the time span is defined as the effective buffer period for clinical operation, the time value of the buffer period is binary encoded and formatted and encapsulated to obtain emergency intervention time window signal data. The remaining time span before the probability density completely returns to zero is calculated, and the critical time point when the probability density value of the curve drops below 0.01 is found. Assuming that the point is 55 minutes, and the current duration is known to be 25 minutes, 55 minutes minus 25 minutes equals 30 minutes. The 30-minute time span is defined as the effective buffer period for clinical operation. That is, if no intervention is performed within 30 minutes, the steady state will completely collapse. The time value of the buffer period (30) is binary encoded and formatted and encapsulated to convert it into a standard 16-bit binary data stream to obtain the emergency intervention time window signal data.

[0037] S503: Call the emergency intervention time window signal data, establish a real-time data transmission channel with the external interactive terminal, map the emergency intervention time window signal to the preset alarm protocol template field, synchronously integrate the risk level code and sound and light drive control parameters, perform instruction-based packaging and integrity verification on the alarm elements, and generate a compensation depletion risk warning instruction. Establish a real-time data transmission channel with external interactive terminals (such as anesthesiologists' handheld PDAs or operating room central control screens), map emergency intervention time window signals to preset alarm protocol template fields, synchronously integrate risk level codes (such as "Level 1 High Risk" corresponding to code 0x01) and sound and light drive control parameters (such as red flashing frequency of 2 Hz, buzzer sound pressure of 85 dB), perform instruction-based packaging and encapsulation of the above alarm elements, and add cyclic redundancy check codes (CRC) for integrity verification, and generate compensation depletion risk warning instructions; Table 3 shows the logic and output content when an alert is triggered: Table 3: Risk Warning Logic Judgment Table Monitoring Projects Real-time values Judgment benchmark / threshold Logical operation result Output Action Steady-state maintenance confidence level 0.368 0.4 Below the benchmark Triggering implicit failure flag Remaining Buffer Period 30 minutes N / A 55-25=30 Encoding input parameters Risk level Level 1 N / A Mapping Drive red audible and visual alarm Table 3 lists the test results of the embodiments. The results show that by comparing the confidence index with the benchmark value of 0.4, it is possible to accurately capture weak risk signals with a value as low as 0.032 and quickly convert them into a visual 30-minute countdown warning, thereby predicting the risk of compensatory exhaustion in advance before physiological indicators show obvious collapse.

[0038] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the technical solution.

Claims

1. A method for intraoperative risk warning in surgical anesthesia based on AI multimodal data fusion, characterized in that, Includes the following steps: S1: Based on the patient's anesthesia medical records, identify continuous stable periods in which arterial pressure and bispectral index of electroencephalogram maintain the standard range, perform statistical analysis on the distribution frequency of stable periods in differentiated time intervals, connect the frequency points to form a probability density shape, and construct an ideal steady-state residence time distribution model; S2: Referring to the normal parameter range defined by the ideal steady-state residence time distribution model, monitor real-time vital signs, track the real-time steady-state duration, and simultaneously collect intraoperative blood loss data and anesthetic drug infusion data. Perform cumulative calculation on intraoperative blood loss and integrate drug concentration at the pharmacokinetic level to generate quantitative indicators of intraoperative cumulative physiological load. S3: Call the intraoperative cumulative physiological load quantification index, calculate the compression ratio coefficient of the distribution area on the time axis based on the load intensity, perform coordinate transformation operation, and obtain the dynamic drift survival probability density curve; S4: Using the dynamic drift survival probability density curve, locate the coordinate point corresponding to the duration of real-time steady state, calculate the coverage area of ​​the time region after the coordinate point, assess the probability that the body will continue to maintain a steady state, and obtain the confidence index of physiological homeostasis maintenance.

2. The method for intraoperative risk warning in surgical anesthesia based on AI multimodal data fusion according to claim 1, characterized in that, The ideal steady-state residence time distribution model includes baseline probability density function parameters, confidence boundaries of the steady-state time interval, and baseline values ​​of the integral area. The quantitative indicators of intraoperative cumulative physiological load include the weighted component of blood loss load, the integral component of drug metabolism residue, and the real-time physiological stress score. The dynamic drift survival probability density curve includes a set of time axis coordinates after compression transformation, the corrected instantaneous survival probability value, and the slope of the dynamic decay trend. The confidence index of physiological homeostasis maintenance includes the residual steady-state probability value under real-time conditions, the normalized percentage of compensatory capacity margin, and the risk rate of homeostasis maintenance failure.

3. The surgical anesthesia intraoperative risk warning method based on AI multimodal data fusion according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Based on the patient's anesthesia medical records, extract the entire arterial pressure monitoring numerical sequence and EEG bispectral index waveform data, call the preset physiological parameter standard maintenance range, compare the overlap between the arterial pressure monitoring numerical sequence and the physiological parameter standard maintenance range one by one, locate the start time stamp and end time stamp when the values ​​of both are maintained within the physiological parameter standard maintenance range, extract the continuous time span between the two time stamps, and obtain the dual-parameter synchronous steady-state time domain segment. S102: Call the dual-parameter synchronous steady-state time domain segment, calculate the segment's time length value, establish a differentiated time interval histogram statistical box, map the time length value to the corresponding histogram statistical box, accumulate the distribution frequency, use the time interval center value as the horizontal axis and the distribution frequency as the vertical axis to draw discrete points, connect adjacent discrete frequency points to form a continuous morphological trajectory, and generate a steady-state duration frequency envelope curve; S103: For the steady-state duration frequency envelope curve, define the closed region enclosed by the curve edge trajectory and the time coordinate axis, perform area calculation on the closed region, obtain the quantitative value characterizing the body's baseline tolerance, use the quantitative value to normalize and correct the amplitude of the curve's vertical axis, determine the probability density function mapping relationship under differentiated durations, and construct an ideal steady-state residence time distribution model.

4. The surgical anesthesia intraoperative risk warning method based on AI multimodal data fusion according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Call the normal parameter range defined by the ideal steady-state residence time distribution model, read the patient's arterial pressure and EEG bifrequency index in real time through the monitoring interface, compare the real-time values ​​with the normal parameter range, determine the steady state and record the duration, generate the real-time steady-state duration, and simultaneously collect the suction bottle fluid volume scale and the surgical gauze weight gain value. Perform cumulative summation on the blood loss data collected in a single session according to the time series to obtain the total cumulative blood loss during the operation. S202: Based on the total cumulative blood loss during the operation, monitor the real-time flow rate setting and drug addition dose of the anesthetic pump, set the drug metabolism attenuation coefficient and compartment distribution volume parameters, perform convolution processing and plasma concentration simulation on the pump data, calculate the residual active drug concentration in the real-time effect compartment, and perform integral integration operation on the cumulative effect of the active concentration over time to obtain integrated pharmacokinetic drug concentration data. S203: Call the integrated pharmacokinetic drug concentration data, set a first weighting coefficient based on the effect of blood loss factors on the loss of circulating volume, set a second weighting coefficient based on the inhibitory intensity of drug factors on the central nervous system, perform weighted multiplication operations on the two data, and perform linear superposition and summation operations on the weighted blood loss value and the weighted integrated drug value to generate a quantitative index of intraoperative cumulative physiological load.

5. The surgical anesthesia intraoperative risk warning method based on AI multimodal data fusion according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Call the intraoperative cumulative physiological load quantification index, retrieve the mapping association list between load intensity and time axis scaling ratio, lock the corresponding time compression ratio coefficient according to the load index value, and simultaneously set the step displacement value of the distribution area moving towards the coordinate origin according to the load magnitude. Perform vectorization combination operation on the compression coefficient and step displacement value to generate time axis compression and translation transformation parameter vector. S302: Call the time axis compression and translation transformation parameter vector, extract the horizontal axis coordinate point set that defines the steady state duration in the ideal steady state residence time distribution model, perform multiplicative shrinkage operation on the horizontal axis coordinate point set using the compression ratio coefficient, and reduce the distribution span of the steady state duration through coordinate transformation operation while keeping the vertical axis probability density amplitude constant, reconstruct the correspondence between the shrunken coordinate point set and the probability density, and generate a time-domain shrinkage distribution morphology dataset; S303: For the time-domain contraction distribution pattern dataset, read the step displacement values ​​in the time axis compression and translation transformation parameter vector, perform subtraction operation on the time coordinate points, implement the overall translation operation towards the coordinate zero point, identify the coordinate points that fall into the negative value interval after translation and perform zero-truncation processing, redraw the distribution envelope edge trajectory according to the updated coordinate point sequence, and obtain the dynamic drift survival probability density curve.

6. The surgical anesthesia intraoperative risk warning method based on AI multimodal data fusion according to claim 5, characterized in that, The mapping association list between load intensity and time axis scaling ratio adopts a discrete interval table structure that sorts the load index value intervals in ascending order. Each load index value interval corresponds to a unique compression ratio coefficient, which is limited to a positive number less than or equal to one. The step displacement value of the distribution area moving towards the origin of the coordinate system is determined by matching the load magnitude with a preset step displacement scale, and the step displacement value is a non-negative real number. Specifically, the vectorization combination operation involves encapsulating the compression ratio coefficient and the step displacement value into the same parameter vector in a fixed-dimensional order. The zero-truncation process is defined as directly assigning zero to time coordinate points that are less than zero after subtraction, keeping the original values ​​of the remaining time coordinate points unchanged, and redrawing the distribution envelope edge trajectory based on the one-to-one correspondence between the time coordinate point sequence after zero-truncation and the original probability density value.

7. The surgical anesthesia intraoperative risk warning method based on AI multimodal data fusion according to claim 5, characterized in that, The specific steps of S4 are as follows: S401: Call the dynamic drift survival probability density curve, map the duration value as the horizontal axis index to the time axis of the curve, anchor the position of the real-time node in the probability distribution, delineate the lag time region from the real-time node to the zero point of the probability density of the curve, perform definite integral operation on the range enclosed by the probability density envelope trajectory and the time axis within the lag time region, and generate the remaining steady-state survival coverage area. S402: For the remaining steady-state coverage area, traverse the complete span of the time axis from zero to the convergence point of the curve edge, extract the probability density amplitude corresponding to the discrete time points within the span, calculate the total amount of the closed geometric region enclosed by the curve edge trajectory and the time coordinate axis, quantify the distribution weight of the probability density throughout the entire life cycle, and generate the total area of ​​the global probability distribution. S403: Call the total area of ​​the global probability distribution to construct the conditional probability evaluation logic. Perform a division operation with the remaining area value as the numerator and the total area value as the denominator to calculate the probability density ratio after the real-time node, quantify the probability that the organism will continue to maintain a stable state under real-time load conditions, and generate the physiological homeostasis maintenance confidence index.

8. The surgical anesthesia intraoperative risk warning method based on AI multimodal data fusion according to claim 7, characterized in that, The duration value is defined as a non-negative time scalar calculated from the zero point of the time axis of the dynamic drift survival probability density curve, and the position of the real-time node in the probability distribution is determined by the one-to-one correspondence between the duration value and the horizontal axis of the dynamic drift survival probability density curve. The lag time region is defined as a closed interval between the horizontal coordinate of the real-time node and the intersection of the dynamic drift survival probability density curve and the time axis, and the probability density zero point is the time coordinate position where the probability density amplitude first equals zero. The definite integral operation is implemented by numerical integration, which accumulates and sums the probability density amplitudes corresponding to discrete time coordinate points within the lag time region. The convergence point of the curve edge is defined as the time coordinate position where the probability density amplitude is continuously lower than a preset threshold and remains unchanged, and the range of the physiological homeostasis maintenance confidence index is limited to a closed interval from zero to one.

9. The method for intraoperative risk warning in surgical anesthesia based on AI multimodal data fusion according to claim 1, characterized in that, The method further includes step S5: S5: Compare the physiological homeostasis maintenance confidence index with the preset safety warning benchmark, identify the target situation where the physiological homeostasis maintenance confidence index is lower than the safety warning benchmark, determine that the patient is in the latent compensatory failure stage, transmit a signal including the intervention time window to the interactive terminal, and generate a compensatory depletion risk warning instruction. The risk warning instruction for the depletion of compensation includes the warning level of implicit compensation failure, the start and end points of the time window for intervention, and the classification description code of the risk status.

10. The method for intraoperative risk warning in surgical anesthesia based on AI multimodal data fusion according to claim 9, characterized in that, The specific steps of S5 are as follows: S501: Call the physiological homeostasis maintenance confidence index, read the preset safety warning benchmark value, compare the physiological homeostasis maintenance confidence index with the safety warning benchmark value, monitor the difference and polarity characteristics between the two in real time, determine the target abnormal situation where the physiological homeostasis maintenance confidence index is lower than the safety warning benchmark, identify the real-time physiological compensation level of the body, mark the physiological compensation level attribute as a latent failure state, and generate a latent compensation failure state judgment identifier; S502: For the implicit compensatory failure state determination identifier, the tail convergence feature of the dynamic drift survival probability density curve is retrieved again, the remaining time span before the probability density completely returns to zero is calculated, the time span is defined as the effective buffer period for clinical operation, the time value of the buffer period is binary encoded and formatted and encapsulated to obtain emergency intervention time window signal data. S503: Call the emergency intervention time window signal data, establish a real-time data transmission channel with the external interactive terminal, map the emergency intervention time window signal to the preset alarm protocol template field, synchronously integrate the risk level code and sound and light drive control parameters, perform instruction-based packaging and integrity verification on the alarm elements, and generate a compensation depletion risk warning instruction.

Images