An intraoperative blood glucose regulation stability intelligent evaluation method and system

Abstract

The application relates to a method and system for intelligent evaluation of blood glucose regulation stability in surgery. The method comprises: time synchronization of multi-modal intraoperative data, construction of a multi-dimensional synchronized time series data set; based on the data set, parallel calculation of dynamic confidence components from multiple dimensions, fusion of the dynamic confidence components into a quantitative blood glucose reading uncertainty parameter by using evidence theory; on this basis, conversion of the original blood glucose data stream into a blood glucose probability distribution data stream with explicit probability distribution representation, and calculation of a blood glucose regulation stability index; according to the index, generation of a comprehensive stability score with a confidence interval. The method can fundamentally solve the problem of dynamic evaluation of blood glucose data reliability in a strong interference and non-steady-state surgical environment, so that the subsequent intelligent stability evaluation is no longer based on the original data which may have serious deviations, thereby significantly improving the accuracy, reliability and clinical decision support value of the evaluation result.

Description

Technical Field

[0001] This invention belongs to the field of perioperative blood glucose management, and in particular relates to an intelligent evaluation method and system for blood glucose regulation stability during surgery. Background Technology

[0002] Perioperative blood glucose management is a crucial aspect of ensuring patient safety and improving prognosis. Both excessively high and low blood glucose levels are closely associated with increased risk of infection, organ dysfunction, and mortality. Therefore, achieving stable regulation and objective evaluation of intraoperative blood glucose levels is of significant clinical importance.

[0003] In traditional techniques, the evaluation of intraoperative glycemic regulation stability mainly relies on the direct calculation of glycemic data sequences obtained from continuous glycemic monitoring or intermittent blood sampling. Typically, the fluctuation of glycemic levels is quantified by statistically analyzing the percentage of time glycemic values ​​within the target range and calculating indicators such as the coefficient of variation of glycemic blood glucose, thereby assessing whether the regulation process is stable.

[0004] However, current evaluation methods have significant problems. The surgical environment itself is highly perturbed and non-steady-state. Various factors, such as drug infusion (e.g., certain analgesics, vasoactive drugs), dramatic changes in the patient's physiological state (e.g., hypotension, hypoperfusion), and procedural factors, can all uncertainly affect the accuracy of blood glucose monitoring data, leading to reading bias or noise. Existing methods, when evaluating stability, generally treat these potentially perturbed data points the same as reliable data, failing to distinguish their inherent differences in reliability. This results in evaluations potentially based on unreliable data, leading to misleading judgments and failing to truly reflect the physiological stability of the patient's blood glucose regulation. Summary of the Invention

[0005] Therefore, it is necessary to provide a method that can identify and integrate the inherent reliability of data to output a stability evaluation result that is resistant to interference, thereby addressing the aforementioned technical problems.

[0006] In a first aspect, this application provides an intelligent evaluation method for intraoperative glycemic regulation stability, including:

[0007] Multimodal intraoperative data were time-synchronized to obtain multidimensional synchronized time-series data, which were then summarized to generate a multidimensional synchronized time-series dataset. The multidimensional synchronized time-series dataset includes core blood glucose measurements, tissue perfusion and oxygenation indicators, drug infusion records, and physiological rhythm indicators.

[0008] Based on a multidimensional synchronous time-series dataset, confidence component indices are calculated, and all confidence component indices are combined to generate a set of confidence component indices. Among them, the confidence component indices include drug interference confidence component indices, physiological interference confidence component indices, physiological consistency confidence component indices, and reference comparison confidence component indices.

[0009] The confidence component index set is dynamically fused to generate probability distribution parameters; the probability distribution parameters are used to characterize the uncertainty of the current blood glucose measurement value.

[0010] Based on probability distribution parameters, the core blood glucose measurement stream is transformed into a blood glucose probability distribution data stream;

[0011] Based on the blood glucose probability distribution data stream, the blood glucose regulation stability index is calculated;

[0012] Based on the stability index of blood glucose regulation, a stability evaluation result is generated; the stability evaluation result includes a comprehensive stability score and a confidence interval for the comprehensive stability score.

[0013] Furthermore, based on the multidimensional synchronous time-series dataset, confidence component indices are calculated, and all confidence component indices are combined to generate a set of confidence component indices, including:

[0014] Based on the drug infusion records in the multidimensional synchronous time-series dataset, the drug concentration at the sensor site at the current moment is obtained, and based on the concentration and the preset concentration-interference intensity mapping function, the drug interference confidence component index is obtained.

[0015] Tissue perfusion and oxygenation indices from a multidimensional synchronous time-series dataset are input into a pre-trained lightweight neural network model, which outputs a physiological interference confidence component index.

[0016] The drug infusion records and physiological rhythm indicators in the multidimensional synchronous time-series dataset are input into the short-term blood glucose trend prediction model, and the predicted range of blood glucose change trend in the near future is output.

[0017] Based on the core blood glucose measurement values ​​in the multidimensional synchronous time series dataset, the degree of agreement between the actual observed blood glucose change values ​​and the predicted range of blood glucose change trends is calculated, and the degree of agreement is determined as the physiological consistency confidence component index.

[0018] When intermittent blood glucose measurements from a reference instrument are present in the multidimensional synchronous time-series dataset, the relative difference between the current continuous blood glucose measurement and the most recent intermittent blood glucose measurement is calculated, and a reference comparison confidence component index is generated based on the magnitude of the relative difference and the time decay function.

[0019] The confidence component indexes are obtained by considering the confidence component indices for combination drug interference, physiological interference, physiological consistency, and reference comparison.

[0020] Furthermore, the confidence component index set is dynamically fused to generate probability distribution parameters, including:

[0021] Based on each confidence component index in the confidence component index set, multiple evidence bodies are constructed using the Dempster-Shafer evidence theory;

[0022] The evidence is iteratively fused to obtain a fusion result, and based on the fusion result, the reliability and likelihood of the data-reliable proposition are calculated.

[0023] Based on reliability and likelihood, the overall confidence score is calculated using the following formula:

[0024]

[0025] in, To calculate the overall confidence score, For reliability, For likelihood;

[0026] Based on the comprehensive confidence score and the nominal measurement noise variance of the continuous blood glucose monitoring device under ideal conditions, the measurement noise variance of continuous blood glucose monitoring is generated, and the measurement noise of continuous blood glucose monitoring is determined as a probability distribution parameter.

[0027] Furthermore, based on the blood glucose probability distribution data stream, indices of blood glucose regulation stability are calculated, including:

[0028] Based on the blood glucose probability distribution data stream, the precision weight of each data point within the preset evaluation time window is calculated; where each data point consists of the mean and variance parameters of blood glucose at the corresponding time.

[0029] Using precision weights, the weighted mean blood glucose and weighted standard deviation of blood glucose within the preset evaluation time window are calculated;

[0030] The weighted blood glucose mean and weighted blood glucose standard deviation are used to calculate the weighted blood glucose coefficient of variation, which is then determined as the first stability index.

[0031] Based on the preset clinical target range and the average blood glucose value at each time point within the preset evaluation time window, a contribution value is assigned to each time point to obtain the time-based contribution value;

[0032] Based on the contribution value at each time point, the weighted target range time percentage is calculated, and the weighted target range time percentage is determined as the second stability index.

[0033] The first and second stability indices are combined to obtain the glycemic regulation stability index.

[0034] Furthermore, based on glycemic regulation stability indicators, stability evaluation results are generated, including:

[0035] Input the first and second stability indices from the glycemic regulation stability indexes into the aggregation function to generate a comprehensive stability score.

[0036] Based on the mean and variance parameters of blood glucose at each time step in the blood glucose probability distribution data stream, a simulated blood glucose trajectory is generated using the Monte Carlo simulation method.

[0037] Calculate the overall stability score for each simulated blood glucose trajectory, and perform statistical analysis on all overall stability scores in conjunction with the preset statistical significance level to generate confidence intervals for each overall stability score;

[0038] By combining the comprehensive stability scores and their corresponding confidence intervals, the stability evaluation results are obtained.

[0039] Secondly, this application also provides an intelligent evaluation system for intraoperative blood glucose regulation stability, comprising:

[0040] The data processing module is used to synchronize multimodal intraoperative data in time to obtain multidimensional synchronized time-series data, and to summarize the multidimensional synchronized time-series data to generate a multidimensional synchronized time-series dataset. The multidimensional synchronized time-series dataset includes core blood glucose measurements, drug infusion records, tissue perfusion and oxygenation indicators, and physiological rhythm indicators.

[0041] The confidence index calculation module is used to calculate confidence component indices based on a multidimensional synchronous time series dataset, and combine all confidence component indices to generate a set of confidence component indices; among which, the confidence component indices include drug interference confidence component indices, physiological interference confidence component indices, physiological consistency confidence component indices, and reference comparison confidence component indices.

[0042] The parameter generation module is used to dynamically fuse the set of confidence component indicators to generate probability distribution parameters; among which, the probability distribution parameters are used to characterize the uncertainty of the current blood glucose measurement value.

[0043] The data conversion module is used to convert the core blood glucose measurement value stream into a blood glucose probability distribution data stream based on probability distribution parameters.

[0044] The stability index generation module is used to calculate the blood glucose regulation stability index based on the blood glucose probability distribution data stream.

[0045] The results generation module is used to generate stability evaluation results based on glycemic regulation stability indicators; the stability evaluation results include a comprehensive stability score and a confidence interval for the comprehensive stability score.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, the at least one program, the code set or instruction set is loaded and executed by the processor to implement any of the intelligent evaluation methods for blood glucose regulation stability during surgery as described in the embodiments of this application.

[0047] Fourthly, this application also provides a computer-readable storage medium storing at least one piece of program code, which is loaded and executed by a processor to implement the intelligent evaluation method for intraoperative blood glucose regulation stability as described in any of the embodiments of this application.

[0048] The aforementioned intelligent evaluation method and system for intraoperative glycemic regulation stability integrates multimodal intraoperative data to construct a multidimensional synchronous time-series dataset. Based on this dataset, dynamic confidence components are calculated in parallel from multiple dimensions, and evidence theory is used to fuse them into quantified glycemic reading uncertainty parameters. Furthermore, the raw glycemic data stream is transformed into a glycemic probability distribution data stream with a clearly defined probability distribution representation, and a glycemic regulation stability index is calculated. Based on this index, a comprehensive stability score with confidence intervals is generated. This method fundamentally solves the challenge of dynamically assessing the reliability of glycemic data in highly perturbed and non-steady-state surgical environments, ensuring that subsequent intelligent stability evaluations are no longer based on potentially severely biased raw data, thereby significantly improving the accuracy, reliability, and clinical decision support value of the evaluation results. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating an intelligent evaluation method for blood glucose regulation stability during surgery, as described in one embodiment.

[0051] Figure 2This is a flowchart illustrating the steps for calculating a blood glucose regulation stability index based on a blood glucose probability distribution data stream in one embodiment.

[0052] Figure 3 This is a schematic diagram of the structure of an intelligent evaluation system for blood glucose regulation stability during surgery, as described in one embodiment. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] In one embodiment, an intelligent evaluation method for blood glucose regulation stability during surgery is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. Figure 1 As shown, in this embodiment, the method includes the following steps:

[0055] Step S101: Time synchronization of multimodal intraoperative data is performed to obtain multidimensional synchronized time-series data, and the multidimensional synchronized time-series data is summarized to generate a multidimensional synchronized time-series dataset; wherein, the multidimensional synchronized time-series dataset includes core blood glucose measurement values, tissue perfusion and oxygenation indicators, drug infusion records and physiological rhythm indicators.

[0056] Time synchronization refers to calibrating the local times of multiple different data sources to a single standard time source, eliminating time discrepancies. Core blood glucose measurements primarily originate from subcutaneous continuous glucose monitoring devices, which provide a high-frequency estimate sequence of interstitial fluid glucose concentration. Tissue perfusion and oxygenation parameters can be obtained from patient monitors, including but not limited to mean arterial pressure (MAP), pulse oxygen saturation (SpO2), and perfusion index (PI). Drug infusion records can be acquired from intelligent infusion pump systems, recording the infusion rates and cumulative doses of vasoactive drugs (such as norepinephrine), drugs that may interfere with CGM readings (such as acetaminophen), and insulin and glucose. Physiological rhythm parameters can be obtained by analyzing electrocardiogram signals and respiratory waveforms, extracting features including time-domain and frequency-domain indicators of heart rate variability (HRV) and respiratory rate. Subcutaneous continuous glucose monitoring (Continuous Glucose)... Blood glucose monitoring (CGM) devices are non-invasive, continuous blood glucose monitoring devices. They use subcutaneously implanted sensors to capture the glucose concentration in interstitial fluid in real time and convert it into blood glucose data, replacing the traditional single-sampling test and enabling dynamic tracking of blood glucose. Patient monitors are electronic devices used in hospitals for real-time, continuous monitoring of patients' vital signs. They are used to promptly capture abnormal changes in vital signs, providing objective data for medical staff to diagnose, treat, and assess the condition, thus reducing the risk of diagnosis and treatment. Intelligent infusion pump systems are devices for precise and intelligent delivery of intravenous drugs / fluids in clinical settings, replacing manual injection to solve the problems of inaccurate drug delivery speed and dosage. Electrocardiograms (ECG) capture the weak current generated by the heart's electrical activity through electrodes on the body surface and convert it into a visual waveform to reflect the electrophysiological state of the heart.

[0057] For example, the system receives heterogeneous data streams in real-time or near real-time from various medical devices and medical information systems, including core blood glucose measurements, tissue perfusion and oxygenation indicators, drug infusion records, and circadian rhythm indicators. It can also receive intermittent, high-precision plasma glucose reference measurements from bedside blood gas analyzers or bedside blood glucose meters. All incoming data packets are timestamped uniformly. For data with different sampling frequencies, a time-based alignment strategy can be employed: high-frequency vital sign data are downsampled to the sampling frequency of CGM data, intermittent reference blood glucose measurements are time-point matched, and all data sequences are linearly interpolated to generate complete observations at uniformly equidistant time points. These processed data streams are then integrated according to time points to generate a structured, multidimensional, synchronous time-series dataset, which includes data points or labels for all the aforementioned dimensions at each sampling time.

[0058] Among them, medical equipment is the hardware carrier that directly serves diagnosis, treatment, monitoring, and rehabilitation; medical information systems are software and hardware integrated systems that process, store, transmit, and analyze medical-related data, used to break down information silos in various medical processes; bedside blood gas analyzers are bedside devices that quickly detect gases, acids, bases, and electrolytes in the blood, emphasizing immediacy, eliminating the need for sending to a central laboratory, and providing results in minutes; bedside blood glucose meters are portable devices that quickly detect capillary blood glucose concentration, emphasizing convenience, low cost, and universality, applicable to almost all clinical departments, and also usable for patients to self-test at home; time-based alignment strategies use the time dimension as the core benchmark to align different data sources... The behavior / data of signals, tasks, or systems are precisely synchronized, matched, or time-series coordinated on the timeline; downsampling is the process of sampling continuous, high-frequency signals or data at certain intervals to reduce their sampling rate or data volume; time point matching refers to accurately mapping time markers from different sources, formats, and precisions to the same time base or target time dimension to achieve time dimension alignment, allowing related data or events to form a correlated and comparable correspondence on the timeline; linear interpolation is a simple interpolation method that uses a straight line to estimate the value of an unknown intermediate point between two known data points, with the core assumption that the variable between the two points changes linearly.

[0059] Step S102: Based on the multidimensional synchronous time series dataset, the confidence component indices are calculated and all confidence component indices are combined to generate a set of confidence component indices; wherein, the confidence component indices include drug interference confidence component indices, physiological interference confidence component indices, physiological consistency confidence component indices, and reference comparison confidence component indices.

[0060] For example, based on drug infusion records, tissue perfusion and oxygenation indicators, physiological rhythm indicators and core blood glucose measurements in a multidimensional synchronous time-series dataset, the confidence component indices for drug interference, physiological interference, physiological consistency and reference comparison are calculated; and the confidence component indices for reference comparison are combined to obtain a set of confidence component indices.

[0061] Step S103: Dynamically fuse the confidence component index set to generate probability distribution parameters; wherein, the probability distribution parameters are used to characterize the uncertainty of the current blood glucose measurement value.

[0062] Dynamic fusion refers to breaking away from static and fixed fusion models and flexibly adjusting the rules, dimensions, weights, or methods of fusion based on real-time changes in scenarios, data, and needs, so that the fusion results can adapt to dynamically changing objectives.

[0063] For example, the Dempster-Shafer (DS) evidence theory framework can be used to fuse multiple independent, potentially mutually supportive or conflicting pieces of confidence evidence into a probability distribution parameter characterizing the uncertainty of the current blood glucose measurement. DS evidence theory is a fusion framework for handling uncertain information. Its core principle is to overcome the limitations of single-element probability assignment in probability theory. It assigns confidence to the power set of a proposition through basic probability assignment, then fuses multi-source evidence using the Dempster synthesis rule, ultimately outputting metrics such as confidence and likelihood of the proposition. The Dempster synthesis rule is a core synthesis method in evidence theory, used to fuse uncertain information from multiple independent evidence sources. It orthogonally sums two basic probability assignment functions to obtain a new basic probability assignment reflecting the synthesized confidence level. This rule eliminates conflicting parts through normalization, meaning different pieces of evidence simultaneously assign positive probability components to mutually exclusive propositions, emphasizing consensus. However, when the conflict between evidence is extremely high, the normalization factor approaches zero, potentially leading to unstable or meaningless results, which is its main limitation.

[0064] Step S104: Based on the probability distribution parameters, the core blood glucose measurement value stream is converted into a blood glucose probability distribution data stream.

[0065] For example, based on probability distribution parameters For the original core blood glucose measurement stream, i.e., the CGM reading sequence This is reformulated, giving it a clear measure of uncertainty. Instead of treating the blood glucose value at each sampling time as a deterministic scalar, it is modeled as a probability distribution. This is achieved by assuming that, given all available multi-source evidence, the true blood glucose value follows a probability distribution based on the current CGM observation. The mean and probability distribution parameters are given. The variance follows a normal (Gaussian) distribution, i.e. For each sampling time The parameters that generate this Gaussian distribution are: mean. and variance Based on this assumption, the original blood glucose value time series... Transformed into a blood glucose probability distribution data stream This data stream provides point estimates of blood glucose levels, and the variance indicates the reliability of these estimates. A larger variance indicates higher uncertainty in the reading at that moment, and the estimated value should be assigned a lower weight in subsequent analysis; conversely, a smaller variance indicates higher reliability. Modeling, in essence, is an abstraction and simplification of the real world. By extracting core elements and clarifying the relationships between these elements, a describable, analyzable, and predictable virtual representation is constructed to solve practical problems. Transformation is the shift from one state, form, property, or subject to another; it is a cross-state or dimensional change, distinct from simple alteration, emphasizing transformations with clear results, goals, or values.

[0066] Step S105: Based on the blood glucose probability distribution data stream, calculate the blood glucose regulation stability index.

[0067] For example, for each time point in the blood glucose probability distribution data stream within a preset evaluation time window, the data is characterized by the mean and variance. A precision weight for that data point is defined based on the reciprocal of the variance. A weighted statistic within the window is calculated based on these precision weights. Based on the obtained weighted statistics, a blood glucose regulation stability index is calculated, including the weighted coefficient of variation of blood glucose and the weighted percentage of time within the target range.

[0068] Step S106: Based on the blood glucose regulation stability index, generate stability evaluation results; wherein, the stability evaluation results include a comprehensive stability score and a confidence interval for the comprehensive stability score.

[0069] For example, the calculated stability indices are combined into an intuitive comprehensive score, and the credibility of the score itself is evaluated to generate a stability evaluation result that includes the comprehensive score and confidence interval.

[0070] In this embodiment, by integrating multimodal physiological and contextual data during surgery, a multidimensional synchronous time-series dataset is constructed. Dynamic confidence components are calculated in parallel from multiple dimensions, including drug interference, physiological state, and signal prediction, and then fused using evidence theory into quantified blood glucose reading uncertainty parameters. Based on this, the raw blood glucose data stream is transformed into a blood glucose probability distribution data stream with a clearly defined probability distribution representation, and a blood glucose regulation stability index is calculated. According to this index, a comprehensive stability score with confidence intervals is generated. This approach fundamentally solves the challenge of dynamically assessing the reliability of blood glucose data in highly interfering and non-steady-state surgical environments. Subsequent intelligent stability evaluations are no longer based on potentially severely biased raw data, but rather on robust data that has undergone uncertainty quantification and weighting, thereby significantly improving the accuracy, reliability, and clinical decision support value of the evaluation results.

[0071] In one embodiment, confidence component indices are calculated based on a multidimensional synchronous time-series dataset, and all confidence component indices are combined to generate a set of confidence component indices, including:

[0072] Step S201: Based on the drug infusion records in the multidimensional synchronous time-series dataset, obtain the drug concentration at the sensor site at the current moment, and obtain the drug interference confidence component index based on the concentration and the preset concentration-interference intensity mapping function.

[0073] The preset concentration-interference intensity mapping function is usually calibrated based on the in vitro cross-reaction experimental data of the sensor (referring to the CGM sensor) and the interfering substance. Its shape is designed as an S-shaped curve or a piecewise linear function, which can map the estimated concentration to an interference intensity coefficient between 0 and 1.

[0074] For example, drug infusion records for a known interfering drug in a multidimensional synchronous time-series dataset are analyzed. These records include the infusion rate, start and end times, and cumulative dose. The plasma concentration of the drug in the blood at the current moment can be simulated using a pharmacokinetic (PK) model based on preset parameters of the known interfering drug and the drug infusion records. Considering the diffusion equilibrium delay between subcutaneous interstitial fluid concentration and plasma concentration, a first-order hysteresis loop can be introduced to estimate the drug concentration near the sensor interface. According to a preset concentration-interference intensity mapping function, the drug concentration is converted into an interference intensity coefficient ranging from 0 to 1. When the concentration is below the no-effect threshold, the interference intensity coefficient is 0; within the therapeutic concentration range, the interference intensity coefficient increases linearly or non-linearly with increasing concentration; at extremely high concentrations, the interference intensity coefficient saturates at its maximum value. The drug interference confidence component index is calculated based on the interference intensity coefficient using the following formula: Among them, known interfering drugs refer to substances known in the dataset that can cause electrochemical interference to the readings of continuous glucose monitoring sensors; The confidence component index for drug interference has a value range of [0,1]. The closer the value is to 1, the lower the current level of drug interference, and the more reliable the CGM reading is in this dimension. The interference intensity coefficient, The maximum value of the interference intensity coefficient; the PK model is a tool that uses mathematical methods to quantitatively describe the changes in drug absorption, distribution, metabolism, and excretion in the body over time. Its core is to transform the dynamic changes in drug concentration in the body into a calculable and predictable mathematical expression; preset parameters refer to distribution volume, clearance rate, or half-life, etc., which are pre-set reference values ​​based on drug development data and population pharmacokinetic characteristics. These are used to assess the drug's disposal patterns in the body and are also an important basis for judging drug interference effects (such as on the metabolism of other drugs); diffusion equilibrium delay refers to the time lag that occurs from the start of diffusion to reaching a spatially uniform concentration equilibrium state during the diffusion process. Essentially, it is caused by the finite rate of diffusion. The establishment of equilibrium is not instantaneous; the first-order lag is a fundamental element in control theory that describes the output lagging behind the input and exhibiting an exponential response. By combining the drug's characteristics with a preset time constant, the kinetic lag characteristics of the drug in vivo can be accurately characterized; the no-effect threshold refers to the minimum effective concentration threshold. When the concentration of a substance does not reach this value, it cannot effectively bind to targets such as receptors and enzymes, nor can it initiate subsequent biological signaling pathways; the therapeutic concentration range refers to the range of blood drug concentrations in the patient's body where the drug achieves an effective therapeutic effect and is relatively safe, i.e., the range between the minimum effective concentration and the maximum safe concentration. Within this range, the drug can exert the expected therapeutic effect while avoiding treatment failure due to excessively low concentrations or toxic reactions due to excessively high concentrations.

[0075] Step S202: Input the tissue perfusion and oxygenation indices from the multidimensional synchronous time-series dataset into a pre-trained lightweight neural network model and output the physiological interference confidence component index.

[0076] The pre-trained lightweight neural network model can be a feedforward neural network with one or two hidden layers. The training data for this model comes from historical intraoperative datasets, which include synchronously recorded physiological indicators, CGM readings, and intermittent arterial blood gas and glucose values ​​considered as relatively accurate references. The training target (label) is not directly the confidence level, but rather the absolute relative error or error level between the CGM reading and the reference blood glucose value. Through supervised learning, the model can automatically learn from the input physiological features the complex nonlinear relationship between states such as hypotension, hypoxia, or peripheral vasoconstriction (manifested as low PI) and CGM sensor signal distortion, delay, or inaccuracy. Related concepts: Feedforward neural networks are the most basic artificial neural network structure, characterized by unidirectional forward data propagation, no loops, and no feedback connections. They form the basis for subsequent convolutional, recurrent, and other neural networks. Supervised learning is one of the core branches of machine learning, characterized by training a model with labeled datasets, allowing the model to learn the mapping relationship between input and output labels, and enabling it to make accurate predictions for new unlabeled inputs after training. The CGM sensor is the core hardware for continuous blood glucose monitoring. It is a small electrochemical sensor that can be minimally invasively attached to the skin to detect the glucose concentration in interstitial fluid in real time and continuously, replacing the single-sampling test of traditional finger-prick blood sampling and realizing dynamic blood glucose monitoring.

[0077] For example, real-time updated tissue perfusion and oxygenation indicators are extracted from a multidimensional synchronous time-series dataset. Typical input features include mean arterial pressure, pulse oxygen saturation, and peripheral perfusion index. These indicators are then input into a pre-trained lightweight neural network model, which directly outputs a scalar value. This scalar value can be normalized to the range of 0 to 1 using an activation function, such as the sigmoid function, to generate a physiological interference confidence component index. The theoretical range is (0,1), which can be described as approximated or regarded as [0,1]. The closer the value is to 1, the better the current tissue perfusion and oxygenation status, which is beneficial for accurate CGM monitoring; the closer it is to 0, the greater the risk of low perfusion or low oxygenation, and the physiological reliability of the CGM reading is questionable. The activation function is the core function in a neural network that introduces nonlinear characteristics into neurons, acting on the weighted input of the neuron, and the output serves as the input to the next layer; the sigmoid function is a classic sigmoid nonlinear activation function used to map any real number to the (0,1) interval; normalization is the core method of data preprocessing, aiming to map feature data of different dimensions and numerical ranges to a unified fixed interval.

[0078] Step S203: Input the drug infusion records and physiological rhythm indicators from the multidimensional synchronous time-series dataset into the short-term blood glucose trend prediction model, and output the predicted range of blood glucose change trend in the future short term.

[0079] Among them, the short-term blood glucose trend prediction model is based on recent blood glucose monitoring data and real-time / short-cycle characteristics affecting blood glucose. It uses algorithms to predict blood glucose values / change trends (rise / fall / stable) for the next few minutes to hours. It is the core technology of dynamic blood glucose management for diabetes. This prediction model can be a well-trained recurrent neural network (RNN). A well-trained recurrent neural network refers to a recurrent neural network model that has completed data training and has fixed parameters. It can use its temporal memory characteristics to complete tasks such as prediction, classification, and generation on sequence inputs.

[0080] For example, based on two types of information in a multidimensional synchronous time-series dataset: first, the insulin and glucose infusion rates in drug infusion records, which are the most direct exogenous factors affecting blood glucose changes; and second, physiological rhythm indicators, such as the frequency domain components (e.g., the low-frequency to high-frequency power ratio) or time domain indicators of heart rate variability extracted from electrocardiogram signals, which reflect autonomic nerve tension and stress levels and are closely related to endogenous glucose metabolism regulation. The short-term blood glucose trend prediction model uses the time series of the aforementioned data as a basis, such as window data from the past tens of minutes, to predict the change in blood glucose within a short future period (e.g., the next 5 minutes). It not only includes point estimates of the predicted change but also outputs the range of uncertainty for the prediction, usually given as the variance of the prediction distribution or the upper and lower limits of the prediction interval. This prediction range reflects the model's confidence in its own prediction results, ultimately resulting in a prediction distribution containing the mean and variance. ,in, The point estimate of the predicted change is used as the predicted mean. This is to predict the variance of the distribution.

[0081] Step S204: Based on the core blood glucose measurement values ​​in the multidimensional synchronous time-series dataset, calculate the degree of agreement between the actual observed blood glucose change values ​​and the predicted range of blood glucose change trends, and determine the degree of agreement as the physiological consistency confidence component index.

[0082] For example, the core blood glucose measurement values, i.e., CGM readings, are obtained from a multidimensional synchronous time-series dataset for the current and previous sampling times. and And according to the formula: Calculate the actual observed changes in blood glucose levels. Predicting distribution based on blood glucose changes within the same time interval. Using probabilistic principles, the degree of agreement is calculated: assessing the likelihood that the observed blood glucose change falls within the predicted distribution. Specifically, the probability density function (PDF) value of the actual observed blood glucose change under the predicted normal distribution is calculated, denoted as... ,Should The value is used to reflect the probability of observing such a magnitude of blood glucose change under the current physiological and therapeutic context. Exactly equal to the predicted mean ,but To obtain the maximum value; if The further the deviation from the predicted mean, the more... The smaller the value, the better. This probability density value is then normalized using the formula: The physiological consistency confidence component index was calculated. The physiological consistency confidence component index has a value range of [0,1]. A higher value indicates that the actual blood glucose change is more consistent with the independent expectation based on other physiological signals, and thus the CGM reading is more reliable in the physiological consistency dimension. The normalization reference value is a preset value, usually corresponding to the maximum probability density under the current prediction variance when the observed value and the predicted mean are completely consistent (i.e., zero deviation). The core idea of ​​probability theory is to use quantitative mathematical methods to describe and analyze the regularity of random phenomena, breaking the perception that randomness is irregular. The core is to find definite probability laws from uncertainty and use them to quantify and predict the possibility of random events. The probability density function value refers to the function value of the probability density function of a continuous random variable at a certain point. This value itself is not a probability, that is, it does not represent the actual probability of the event occurring, but reflects the relative likelihood of the random variable taking a value near that point. Its value can be greater than 1, but the integral of the probability density function over the entire real number axis is equal to 1.

[0083] Step S205: When there are intermittent blood glucose measurements from a reference instrument in the multidimensional synchronous time-series dataset, calculate the relative difference between the current continuous blood glucose measurement and the most recent intermittent blood glucose measurement, and generate a reference comparison confidence component index based on the magnitude of the relative difference and the time decay function.

[0084] Reference instruments refer to bedside testing devices or laboratory analytical instruments used during surgery for discrete, intermittent blood glucose measurements, which typically have high accuracy and clinical acceptance, such as blood gas analyzers and bedside blood glucose meters.

[0085] For example, by continuously monitoring a multidimensional synchronous time-series dataset, this calculation is triggered when a new intermittent blood glucose measurement value and its corresponding timestamp from a reference instrument are detected. The calculation retrieves the CGM reading closest to the reference measurement time and calculates the absolute relative difference (ARD) between the two. The formula for calculation is as follows: .in, For absolute and relative differences, The CGM reading that is closest to the reference measurement time. Intermittent blood glucose measurements from a reference instrument. The reference measurement time is the timestamp corresponding to the intermittent blood glucose measurements from the reference instrument. For example, based on the magnitude of this difference, an initial comparison confidence level at the reference measurement time is determined using a pre-defined lookup table. Since the reference measurements are sparse, their calibration effect weakens over time. This effect can be simulated by introducing a time decay function, which typically takes an exponential decay form. For the current assessment time (in The reference comparison confidence component index is calculated by multiplying the initial confidence level obtained from the most recent reference comparison by a decay factor, i.e. .in, The reference comparison confidence component index is reset and updated each time a new reference value is obtained, and is used to provide a reliability measure that decays over time based on the gold standard comparison. Initial comparison confidence level; The time constant is a preset parameter used to reflect the rate at which the validity of the reference value decays. It can be set based on clinical experience or statistical analysis. The preset lookup table refers to a set of rules defined in advance. Essentially, it is a fixed correspondence between input and output. It does not require real-time calculation; it can be directly looked up. The mapping principle is: the smaller the ARD is, the closer the initial confidence level is to 1; when the ARD exceeds a certain acceptable error threshold, the initial confidence level tends to be close to 0.

[0086] Step S206: Combine the confidence component indices of drug interference, physiological interference, physiological consistency, and reference comparison to obtain a set of confidence component indices.

[0087] Among them, the confidence component of drug interference, the confidence component of physiological interference, the confidence component of physiological consistency, and the confidence component of reference comparison are used to characterize the different types of reliability risks or assurances faced by CGM readings at a specific moment from four independent and complementary perspectives: pharmacology, physiological state, trend rationality, and external calibration.

[0088] For example, the confidence components for drug interference, physiological interference, physiological consistency, and reference comparison calculated at the current moment are extracted. These indicators are all scalar values ​​defined in the interval [0,1]. These scalar values ​​can be organized in a preset order to generate a structured dataset, namely the set of confidence component indicators. ,For example The preset order refers to a pre-defined sequence with fixed logic or rules.

[0089] In this embodiment, by quantitatively characterizing the deterministic interference risk caused by specific drug infusions, a neural network is used to intelligently assess the complex physiological interferences caused by the patient's circulatory and oxygenation status. An independent prediction model based on non-glucose physiological signals is established to verify the physiological rationality of CGM reading changes. High-precision sparse reference measurements are introduced for external calibration and timeliness verification. All independent assessment results are integrated to obtain a set of confidence component indices. This systematically transforms various elusive interference factors in the surgical environment into a series of calculable and quantifiable dynamic confidence component indices, thus laying a fine and solid evidentiary foundation for subsequent fusion to generate a global uncertainty metric.

[0090] In one embodiment, the confidence component index set is dynamically fused to generate probability distribution parameters, including:

[0091] Step S301: Based on each confidence component index in the confidence component index set, construct multiple evidence bodies using the Dempster-Shafer evidence theory.

[0092] For example, define the recognition framework ,in, This implies that blood glucose data is reliable. The propositions representing the unreliability of blood glucose data are mutually exclusive and exhaust all possibilities. For example, for each confidence component in the set of confidence component indicators... Construct a Basic Probability Assignment (BPA) function for it. This function assigns probability quality to the recognition frame. A subset of. In a typical and efficient implementation, a deterministic allocation strategy can be adopted: assigning index values... Directly assigned to a single-point subset This indicates the quality of supporting the proposition that the data is reliable; Assigned to a single-point subset This indicates the quality of supporting the proposition that data is unreliable; while the identification framework itself... The quality of the assignment (i.e., the unknown state) is set to 0. This can be expressed mathematically as: , , Each confidence component index processed in this way, along with its BPA function, constitutes an independent body of evidence. This process is repeated for all indicators in the set, thereby outputting a set of bodies of evidence that correspond one-to-one with the input indicators. .

[0093] Step S302: Iteratively fuse the various evidence bodies to obtain the fusion result, and calculate the reliability and likelihood of the data-reliable proposition based on the fusion result.

[0094] Iterative fusion refers to the process of gradually integrating, verifying, and optimizing multiple independent sources of evidence (evidence bodies) through repeated iterations, ultimately forming a unified and reliable comprehensive evidentiary conclusion.

[0095] For example, the Dempster combination rule can be used for iterative fusion. This rule is used to merge two pieces of evidence; for more pieces of evidence, it is achieved through pairwise iteration. For any two pieces of evidence... and Its combination rule is defined as: ,in, To identify the frame For any subset of a given set, summate all such subsets. and The intersection is of ; This is called the conflict coefficient, used to quantify the degree of conflict between two pieces of evidence. This serves a normalization function, redistributing the quality of conflict. For example, starting with the first piece of evidence, it is sequentially combined with subsequent pieces of evidence. After fusing all pieces of evidence, the final comprehensive basic probability allocation function is obtained. Based on the comprehensive basic probability assignment function, the reliability and likelihood of the proposition data are calculated. Among them, the reliability... Defined as the sum of the basic probability masses of all subsets that support the proposition, which in this case is the single-point subset. The sum of their basic probability masses, i.e. Likelihood Defined as the sum of the basic probability qualities of all subsets that do not conflict with the reliable propositions of the data, i.e., containing A subset of, including and The sum of the basic probability masses, i.e. Reliability represents the minimum level of support for a proposition, while likelihood represents the maximum level of support for a proposition. Together, they constitute a confidence interval. Combinatorial operations are the core operations of combinatorics. They are used to calculate the number of all possible choices of k elements from n distinct elements, regardless of their order. They are the unordered version of permutation operations. k can be chosen according to the actual situation.

[0096] Step S303: Based on the reliability and likelihood, calculate the overall confidence score using the following formula:

[0097]

[0098] in, To calculate the overall confidence score, For reliability, This is the likelihood.

[0099] For example, after obtaining the fused reliability and likelihood, the formula is used: Calculate the overall confidence score The formula is calculated based on the confidence interval formed by the reliability and likelihood. This can fully express the range of support for the proposition that the data is reliable given all existing evidence. Taking the midpoint of this interval as the point estimate is a robust strategy that balances the most conservative (reliability) and the most optimistic (likelihood) viewpoints. Since both reliability and likelihood have a range of [0,1] and satisfy... Therefore, the calculated overall confidence score The value range is also [0,1]. When all evidence highly consistently supports the reliability of the data, and Both are close to 1. Both are close to 1; when all evidence highly consistent against the reliability of the data, both are close to 0. It is also close to 0; when there is significant conflict or uncertainty among the evidence, the interval widens, and the midpoint... This is used to reflect a compromise assessment under such uncertainty.

[0100] Step S304: Based on the comprehensive confidence score and the nominal measurement noise variance of the continuous blood glucose monitoring device under ideal conditions, generate the continuous blood glucose monitoring measurement noise variance and determine the continuous blood glucose monitoring measurement noise as a probability distribution parameter.

[0101] The nominal measurement noise variance is usually provided by the manufacturer of the continuous glucose monitoring device or obtained through prior calibration experiments. It is used to characterize the inherent measurement error level of the sensor model under interference-free and physiologically stable conditions, and serves as a basic reference value.

[0102] For example, the nominal measurement noise variance of a continuous glucose monitoring (CGM) device under ideal conditions is obtained, denoted as . Establish and apply a mapping function that integrates confidence scores. As input, the output is the dynamic measurement noise variance. Its form of expression is: .in, For a very small positive number (e.g.) ), used to prevent when Theoretically, when the overall confidence score is 0, the denominator is zero. The design principle of this function is: when the overall confidence score is 0... When it is very high (close to 1), This is used to indicate that measurement uncertainty is basically determined by the inherent performance of the sensor; when When the value decreases, the denominator decreases. An increase means that due to various interferences, the certainty of the current reading decreases, and its uncertainty is amplified; when When it approaches 0, It tends towards a very large value, which is mathematically equivalent to the data point providing almost no useful information. For example, the calculated... This is the final generated continuous blood glucose monitoring measurement noise variance, which is directly determined as the core probability distribution parameter used to characterize the uncertainty of the current blood glucose measurement value. In subsequent blood glucose probability distribution modeling, this... It will be used as the variance parameter of the normal distribution (or other selected distribution).

[0103] In this embodiment, the multidimensional confidence index is formalized into an evidence body under Dempster's evidence theory; the Dempster combination rule is used to iteratively fuse multi-source evidence to generate confidence intervals representing the range of beliefs; the interval information is condensed into a single comprehensive confidence score; and the comprehensive confidence score is mapped to dynamic measurement noise variance, thus concretizing the abstract reliability assessment into key distribution parameters that can be embedded in a probability model. This ensures that the process from multidimensional qualitative assessment to the generation of a single quantitative parameter retains the richness of evidence while possessing mathematical rigor and computational feasibility, laying a solid foundation for the final generation of a blood glucose probability distribution data stream with precise uncertainty measurement.

[0104] In one embodiment, such as Figure 2 As shown, based on the blood glucose probability distribution data stream, the blood glucose regulation stability index is calculated, including:

[0105] Step S401: Based on the blood glucose probability distribution data stream, calculate the precision weight of each data point within the preset evaluation time window; wherein, the data point consists of the mean and variance parameters of blood glucose at the corresponding time.

[0106] The preset evaluation time window is a fixed time range defined in advance for conducting the evaluation work. This ensures a unified time scale for the evaluation and avoids deviations in results due to ambiguous time definitions. The blood glucose probability distribution data stream within the preset evaluation time window contains a series of data points arranged by timestamps. Each data point is a binary tuple representing the mean blood glucose level at the corresponding time. (Taken from CGM readings) and variance parameter characterizing its uncertainty Composition, that is, data points can be represented as The principle of precision weighting is as follows: the larger the variance of an observation, the higher its uncertainty and the less reliable information it contains. Therefore, it should be given a lower weight in subsequent statistical summaries. Conversely, the smaller the variance, the higher the weight.

[0107] For example, based on the blood glucose probability distribution data stream, and according to the variance-based weighting method in parameter estimation theory, the reciprocal of the variance can be used as the formula for calculating the precision weight. For the first... Data points Its precision weight is given by the formula The calculation yielded the following result. For precision weights, Variance is used to directly determine the magnitude of the weights. For example, to ensure the stability of numerical calculations, in actual calculations, [variance can be used]. A very small positive lower bound is applied to prevent overflow in weight calculation when the variance is zero or close to zero. The precision weight is calculated according to the formula for each data point within the evaluation time window, ultimately generating a precision weight sequence that corresponds one-to-one with each data point within the evaluation time window. .

[0108] Step S402: Using precision weights, calculate the weighted mean blood glucose and weighted standard deviation of blood glucose within the preset evaluation time window.

[0109] For example, the total weight of all data points within a preset evaluation time window is calculated using the following formula: Calculate the weighted mean blood glucose level, which is the weighted average of all blood glucose means with their respective precision weights as coefficients. The formula is as follows: The formula for calculating the weighted variance is: This formula calculates the weighted average of the squared deviations of each blood glucose mean from the weighted average. Taking the square root of this weighted variance yields... This yields the standard deviation with the same dimensions as the original blood glucose value, i.e., the weighted blood glucose standard deviation. For the total weight, The total number of data points within the preset evaluation time window. This is the weighted mean blood glucose level, which represents the best estimate of the average blood glucose level within the window after considering the different confidence levels at each point. This is a weighted variance, which is used to quantify blood glucose fluctuations under the influence of weights; The weighted standard deviation of blood glucose; For precision weights, This represents the average blood glucose level.

[0110] Step S403: Based on the weighted mean blood glucose and the weighted standard deviation of blood glucose, the weighted coefficient of variation of blood glucose is calculated, and the weighted coefficient of variation of blood glucose is determined as the first stability index.

[0111] The weighted coefficient of variation (CV) is an innovation based on the traditional coefficient of variation. The traditional CV is a dimensionless indicator used in statistics to measure the dispersion of data relative to its average level. It is widely used in the field of blood glucose management to assess the stability of blood glucose. The smaller the value, the smoother the blood glucose fluctuation. The traditional CV calculation directly uses the sample mean and sample standard deviation of the original blood glucose values. However, the sample mean and sample standard deviation used by the weighted CV are the results after adjustment by precision weight. Therefore, the weighted CV is not the fluctuation after a simple average of all data points, but the fluctuation calculated mainly based on high confidence data points.

[0112] For example, the weighted coefficient of variation for blood glucose is calculated using the following formula: This weighted coefficient of variation of blood glucose was determined as the first core indicator for evaluating the stability of blood glucose regulation. This is the weighted coefficient of variation for blood glucose. For weighted blood glucose standard deviation, This is the weighted average blood glucose level.

[0113] Step S404: Based on the preset clinical target range and the average blood glucose value at each time point within the preset evaluation time window, assign a contribution value to each time point to obtain the time contribution value.

[0114] Among them, the preset clinical target range refers to the blood glucose control benchmark set in advance based on the patient. This range usually has a lower limit. (Lower Bound) and an upper bound. (Upper Bound, previous definition) is, for example, 4.0 mmol / L to 10.0 mmol / L.

[0115] For example, based on a preset clinical target range and the average blood glucose level at each time point within a preset evaluation time window. and corresponding precision weights For each moment A contribution value is calculated, and the allocation rule for the contribution value is based on an indicator function. The function outputs 1 when the condition is true, and 0 otherwise. Specifically, for each time step... Execute the judgment: If the average blood glucose level at that moment is... satisfy That is, if it falls within the target range, the contribution value at that moment. Set its precision weight ;like Below or higher The contribution value at that moment Set to 0. Its formula is expressed as: Iterate through all moments within the preset evaluation time window to generate a sequence of contribution values. The indicator function, also called the indicative function, is used to mark whether a condition is true or false using the numerical values ​​1 and 0, transforming logical judgments into calculable numerical values. It is a common tool in mathematics, probability theory, and statistics for simplifying expressions. The physical meaning of the allocation rule is that the contribution of a given moment to the target time depends not only on whether the blood glucose level meets the target, but also on the reliability of the reading itself, determined by weights. This demonstrates that a high-confidence benchmark contributes significantly, a low-confidence benchmark contributes little, and any non-qualified benchmark contributes nothing, regardless of confidence level.

[0116] Step S405: Based on the contribution value at each time point, calculate the weighted target range time percentage and determine the weighted target range time percentage as the second stability index.

[0117] For example, based on the time-time contribution value sequence And precision weight sequence Calculate the weighted percentage of time within the target range: calculate the sum of the contribution values ​​at all times within the preset evaluation time window. This sum represents the cumulative amount of all valid achievements; the total weight is calculated for all times within this window. This total weight represents the cumulative amount of all valid monitoring time, because the monitoring behavior at each moment is weighted to represent its information value; the weighted percentage of time within the target range is calculated. The calculation formula is as follows: . This The value was determined as the second core indicator for evaluating the stability of blood glucose regulation.

[0118] Step S406: Summarize the first stability index and the second stability index to obtain the blood glucose regulation stability index.

[0119] For example, the first stability index and the second stability index are combined to obtain the blood glucose regulation stability index. Here, "combining" refers to collecting, organizing, and merging scattered information, data, and content to form a complete and systematic whole.

[0120] In this embodiment, dynamic uncertainty is quantified into precision weights; weighted mean and standard deviation are calculated based on these precision weights; a weighted coefficient of variation that can filter out noise fluctuations is generated based on this; by defining the calculation rules for time within the weighted target range, data reliability is embedded in the achievement rate assessment to obtain the true achievement ratio that can resist misleading interference readings; these two weighted indicators are integrated to form a comprehensive evaluation. This ensures that the stability indicators of the final output are no longer easily affected by brief strong interferences or data quality fluctuations during surgery, improving the accuracy and safety of intraoperative blood glucose management decisions.

[0121] In one embodiment, a stability evaluation result is generated based on a glycemic regulation stability index, including:

[0122] Step S501: Input the first and second stability indices from the blood glucose regulation stability indexes into the aggregation function to generate a comprehensive stability score.

[0123] Aggregate functions are functions that perform aggregate calculations on a set of data and ultimately return a single value. Aggregate calculation refers to the method of integrating and summarizing a set of scattered, multi-dimensional raw data through specific calculation rules, and finally outputting a single or a small number of generalized and statistical results. A typical and interpretable form of aggregate function is a linear weighted sum: , The overall stability score to be generated; and The preset weighting coefficients are pre-defined values ​​that reflect the relative importance clinical patients place on achieving target blood glucose levels and maintaining stable blood glucose fluctuations, satisfying... ; and For example, a normalization function ,like Expressed as a percentage, or a percentage Mapped to an sigmoid function with scores of 0-1; Then it might be a general The reason for the decreasing function of the value inverse mapping is... The smaller the value, the better the stability, for example , This parameter is used for manual / parameter-tuning control of attenuation intensity. It has no unit and only serves a scaling function.

[0124] For example, based on the first and second stability indices in the glycemic regulation stability index, a typical aggregation function form of linear weighted summation can be used. The overall stability score is calculated, which is within a predetermined range, such as 0 to 100. The higher the score, the better the overall stability of blood glucose regulation.

[0125] Step S502: Based on the mean and variance parameters of blood glucose at each moment in the blood glucose probability distribution data stream, a simulated blood glucose trajectory is generated using the Monte Carlo simulation method.

[0126] Monte Carlo simulation is a numerical computation method based on random sampling and statistical analysis. Its core is to approximate the real probability distribution and numerical solution with the results of a large number of random experiments. In essence, it transforms complex deterministic problems into probabilistic problems to be solved. It is named after the random characteristics of the Monte Carlo casino.

[0127] For example, within a preset evaluation time window, the blood glucose probability distribution data stream provides data for each sampling time. mean blood glucose and variance parameter Both are used to define a normal distribution. This represents the probabilistic belief in the actual blood glucose level at that moment. A relatively large number of simulations is pre-set. ,For example For each simulation ( It independently iterates through each moment within the preset evaluation time window. From its corresponding normal distribution Randomly select a sample value Sample values ​​at all times Connecting them in chronological order forms the first... simulated blood glucose trajectory This trajectory represents one possible temporal evolution of true blood glucose levels given the current level of uncertainty. Repeat this process. Next, generate a containing A set of simulated blood glucose trajectories Where N is the number of moments within the preset evaluation time window; random sampling refers to a selection method in which the probability of each individual being selected is objectively equal and free from subjective human intervention when selecting a subset of individuals from a given population (the set of all candidates). The core is to eliminate bias and ensure the randomness of the result. Simply put, it means that there is no pattern or selection when selecting, and each object has an equal chance of being selected. For example, drawing lots, lottery, and random number selection are all common forms of random sampling.

[0128] Step S503: Calculate the overall stability score of each simulated blood glucose trajectory, and perform statistical analysis on all overall stability scores in conjunction with the preset statistical significance level to generate the confidence interval of each overall stability score.

[0129] Among them, the preset statistical significance level, commonly represented by α, is a probability threshold set in advance before hypothesis testing. It is used to define the standard for low-probability events and is the maximum probability allowed to commit a Type I error (false positive). Simply put, it is the critical value for judging whether the result is caused by accidental factors. Statistical analysis refers to the process of collecting, organizing, analyzing, and interpreting data using statistical methods and tools, extracting patterns, discovering trends, and verifying hypotheses from massive and scattered data, and ultimately providing objective data support for decision-making.

[0130] For example, for each simulated blood glucose trajectory, it is treated as a set of possible real blood glucose observations, and a stability index is recalculated for it. Specifically, for the simulated blood glucose trajectory... Calculate its blood glucose variation coefficient directly and percentage of time within the target range It should be noted that no weights are needed at this stage because the simulated values ​​are treated as deterministic observations. The same aggregation function as in step S501 can be used to... and As input, the comprehensive stability score corresponding to the simulated trajectory is calculated. For all This calculation is repeated for each simulated trajectory to obtain a sequence containing... A sample set of comprehensive stability scores This sample set is used to reflect the possible distribution of the final stability evaluation results, taking into account the uncertainty of the original blood glucose data. This is based on a pre-defined statistical significance level. ,For example At a 95% confidence level, perform statistical analysis on this scoring sample set. Calculate the specific percentile of this sample. For example, for a 95% confidence interval, calculate the 2.5th percentile. and the 97.5th percentile Generate a comprehensive stability score confidence interval This interval is used to quantitatively express the range of certainty regarding the final stability score based on currently uncertain data. , .

[0131] Step S504: Combine the comprehensive stability scores and corresponding confidence intervals to obtain the stability evaluation results.

[0132] For example, the comprehensive stability score And the confidence interval of the score This results are combined into a structured stability assessment object. Further information can be added to this result based on clinical interpretation needs, such as information based on scores. The qualitative classification label given by the threshold range, such as stable, critical, unstable, or based on the width of the confidence interval ( The data quality indicators provided include a high degree of certainty in the evaluation. The threshold range is a state interval defined around a critical threshold. The core concept is to use "stable-critical-unstable" to define the position of the indicator value relative to the critical value, and the corresponding operating state of the system / thing. These three are a continuous state gradient, not an isolated classification.

[0133] In this embodiment, discrete stability dimensions are fused into an intuitive comprehensive score using an aggregation function; multiple simulated blood glucose trajectories are generated based on the probability distribution data stream produced in previous steps through Monte Carlo simulation; statistical analysis is performed on a large number of simulation results to quantify the probability distribution of the comprehensive score values ​​and generate their confidence intervals; the point estimates and confidence intervals are combined into a complete evaluation result. This effectively enhances the scientific rigor and clinical credibility of the evaluation results, significantly improving the safety and reliability of clinical decision-making based on this evaluation.

[0134] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0135] Based on the same inventive concept, this application also provides an intelligent evaluation system for intraoperative blood glucose regulation stability to implement the aforementioned intelligent evaluation method for intraoperative blood glucose regulation stability. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the intelligent evaluation system for intraoperative blood glucose regulation stability provided below can be found in the limitations of the intelligent evaluation method for intraoperative blood glucose regulation stability described above, and will not be repeated here.

[0136] In one exemplary embodiment, such as Figure 3 As shown, an intelligent evaluation system 300 for intraoperative blood glucose regulation stability is provided, comprising:

[0137] The data processing module 301 is used to synchronize the multimodal intraoperative data in time to obtain multidimensional synchronized time-series data, and to summarize the multidimensional synchronized time-series data to generate a multidimensional synchronized time-series dataset; wherein, the multidimensional synchronized time-series dataset includes core blood glucose measurement values, tissue perfusion and oxygenation indicators, drug infusion records and physiological rhythm indicators.

[0138] The confidence index calculation module 302 is used to calculate the confidence component index based on the multidimensional synchronous time series dataset, and combine all the confidence component indexes to generate a set of confidence component indexes; among which, the confidence component indexes include drug interference confidence component index, physiological interference confidence component index, physiological consistency confidence component index and reference comparison confidence component index.

[0139] The parameter generation module 303 is used to dynamically fuse the set of confidence component indicators to generate probability distribution parameters; wherein, the probability distribution parameters are used to characterize the uncertainty of the current blood glucose measurement value.

[0140] The data conversion module 304 is used to convert the core blood glucose measurement value stream into a blood glucose probability distribution data stream based on probability distribution parameters.

[0141] The stability index generation module 305 is used to calculate the blood glucose regulation stability index based on the blood glucose probability distribution data stream.

[0142] The result generation module 306 is used to generate stability evaluation results based on blood glucose regulation stability indicators; wherein, the stability evaluation results include a comprehensive stability score and a confidence interval for the comprehensive stability score.

[0143] In one embodiment, the confidence index calculation module 302 is further configured to:

[0144] Based on the drug infusion records in the multidimensional synchronous time-series dataset, the drug concentration at the sensor site at the current moment is obtained, and based on the concentration and the preset concentration-interference intensity mapping function, the drug interference confidence component index is obtained.

[0145] Tissue perfusion and oxygenation indices from a multidimensional synchronous time-series dataset are input into a pre-trained lightweight neural network model, which outputs a physiological interference confidence component index.

[0146] The drug infusion records and physiological rhythm indicators in the multidimensional synchronous time-series dataset are input into the short-term blood glucose trend prediction model, and the predicted range of blood glucose change trend in the near future is output.

[0147] Based on the core blood glucose measurement values ​​in the multidimensional synchronous time series dataset, the degree of agreement between the actual observed blood glucose change values ​​and the predicted range of blood glucose change trends is calculated, and the degree of agreement is determined as the physiological consistency confidence component index.

[0148] When intermittent blood glucose measurements from a reference instrument are present in the multidimensional synchronous time-series dataset, the relative difference between the current continuous blood glucose measurement and the most recent intermittent blood glucose measurement is calculated, and a reference comparison confidence component index is generated based on the magnitude of the relative difference and the time decay function.

[0149] The confidence component indexes are obtained by considering the confidence component indices for combination drug interference, physiological interference, physiological consistency, and reference comparison.

[0150] In one embodiment, the parameter generation module 303 is further configured to:

[0151] Based on each confidence component index in the confidence component index set, multiple evidence bodies are constructed using the Dempster-Shafer evidence theory;

[0152] The evidence is iteratively fused to obtain a fusion result, and based on the fusion result, the reliability and likelihood of the data-reliable proposition are calculated.

[0153] Based on reliability and likelihood, the overall confidence score is calculated using the following formula:

[0154]

[0155] in, To calculate the overall confidence score, For reliability, For likelihood;

[0156] Based on the comprehensive confidence score and the nominal measurement noise variance of the continuous blood glucose monitoring device under ideal conditions, the measurement noise variance of continuous blood glucose monitoring is generated, and the measurement noise of continuous blood glucose monitoring is determined as a probability distribution parameter.

[0157] In one embodiment, the stability index generation module 305 is further configured to:

[0158] Based on the blood glucose probability distribution data stream, the precision weight of each data point within the preset evaluation time window is calculated; where each data point consists of the mean and variance parameters of blood glucose at the corresponding time.

[0159] Using precision weights, the weighted mean blood glucose and weighted standard deviation of blood glucose within the preset evaluation time window are calculated;

[0160] The weighted blood glucose mean and weighted blood glucose standard deviation are used to calculate the weighted blood glucose coefficient of variation, which is then determined as the first stability index.

[0161] Based on the preset clinical target range and the average blood glucose value at each time point within the preset evaluation time window, a contribution value is assigned to each time point to obtain the time-based contribution value;

[0162] Based on the contribution value at each time point, the weighted target range time percentage is calculated, and the weighted target range time percentage is determined as the second stability index.

[0163] The first and second stability indices are combined to obtain the glycemic regulation stability index.

[0164] In one embodiment, the result generation module 306 is further configured to:

[0165] Input the first and second stability indices from the glycemic regulation stability indexes into the aggregation function to generate a comprehensive stability score.

[0166] Based on the mean and variance parameters of blood glucose at each time step in the blood glucose probability distribution data stream, a simulated blood glucose trajectory is generated using the Monte Carlo simulation method.

[0167] Calculate the overall stability score for each simulated blood glucose trajectory, and perform statistical analysis on all overall stability scores in conjunction with the preset statistical significance level to generate confidence intervals for each overall stability score;

[0168] By combining the comprehensive stability scores and their corresponding confidence intervals, the stability evaluation results are obtained.

[0169] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of intelligent evaluation of intraoperative blood glucose regulation stability as described above.

[0170] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0171] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0172] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. An intelligent evaluation method for blood glucose regulation stability during surgery, characterized in that, The method includes: Multimodal intraoperative data is time-synchronized to obtain multidimensional synchronized time-series data, and the multidimensional synchronized time-series data is summarized to generate a multidimensional synchronized time-series dataset; wherein, the multidimensional synchronized time-series dataset includes core blood glucose measurement values, tissue perfusion and oxygenation indicators, drug infusion records and physiological rhythm indicators; Based on the multidimensional synchronous time-series dataset, confidence component indices are calculated, and all the confidence component indices are combined to generate a set of confidence component indices; wherein, the confidence component indices include drug interference confidence component indices, physiological interference confidence component indices, physiological consistency confidence component indices, and reference comparison confidence component indices. The confidence component index set is dynamically fused to generate probability distribution parameters; wherein, the probability distribution parameters are used to characterize the uncertainty of the current blood glucose measurement value; Based on the probability distribution parameters, the core blood glucose measurement value stream is transformed into a blood glucose probability distribution data stream; Based on the blood glucose probability distribution data stream, a blood glucose regulation stability index is calculated; Based on the blood glucose regulation stability index, a stability evaluation result is generated; wherein, the stability evaluation result includes a comprehensive stability score and a confidence interval of the comprehensive stability score.

2. The method according to claim 1, characterized in that, Based on the multidimensional synchronous time-series dataset, confidence component indices are calculated, and all the confidence component indices are combined to generate a set of confidence component indices, including: Based on the drug infusion records in the multidimensional synchronous time-series dataset, the drug concentration at the sensor site at the current moment is obtained, and based on the concentration and a preset concentration-interference intensity mapping function, the drug interference confidence component index is obtained. The tissue perfusion and oxygenation indices in the multidimensional synchronous time-series dataset are input into a pre-trained lightweight neural network model, which outputs a physiological interference confidence component index. The drug infusion records and physiological rhythm indicators in the multidimensional synchronous time-series dataset are input into the short-term blood glucose trend prediction model, and the predicted range of blood glucose change trend in the future short term is output. Based on the core blood glucose measurement values ​​in the multidimensional synchronous time-series dataset, the degree of agreement between the actual observed blood glucose change values ​​and the predicted range of blood glucose change trends is calculated, and the degree of agreement is determined as the physiological consistency confidence component index. When intermittent blood glucose measurements from a reference instrument are present in the multidimensional synchronous time-series dataset, the relative difference between the current continuous blood glucose measurement and the most recent intermittent blood glucose measurement is calculated, and a reference comparison confidence component index is generated based on the magnitude of the relative difference and in conjunction with the time decay function. The confidence component index is obtained by combining the drug interference confidence component index, the physiological interference confidence component index, the physiological consistency confidence component index, and the reference comparison confidence component index.

3. The method according to claim 1, characterized in that, The step of dynamically fusing the set of confidence component indicators to generate probability distribution parameters includes: Based on each confidence component index in the set of confidence component indices, multiple evidence bodies are constructed using the Dempster-Shafer evidence theory; The evidence bodies are iteratively fused to obtain a fusion result, and based on the fusion result, the reliability and likelihood of the data reliability proposition are calculated. Based on the reliability and the likelihood, the overall confidence score is calculated using the following formula: in, The overall confidence score is [value]. For the aforementioned reliability, The likelihood is given. Based on the comprehensive confidence score and the nominal measurement noise variance of the continuous blood glucose monitoring device under ideal conditions, the continuous blood glucose monitoring measurement noise is generated, and the continuous blood glucose monitoring measurement noise is determined as the probability distribution parameter.

4. The method according to claim 1, characterized in that, The blood glucose regulation stability index calculated based on the blood glucose probability distribution data stream includes: Based on the blood glucose probability distribution data stream, the precision weight of each data point within the preset evaluation time window is calculated; wherein, the data point consists of the mean and variance parameters of blood glucose at the corresponding time. Using the precision weights, the weighted mean blood glucose and weighted standard deviation of blood glucose within the preset evaluation time window are calculated; Based on the weighted mean blood glucose and the weighted standard deviation of blood glucose, the weighted coefficient of variation of blood glucose is calculated, and the weighted coefficient of variation of blood glucose is determined as the first stability index; Based on the preset clinical target range and the average blood glucose value at each time point within the preset evaluation time window, a contribution value is assigned to each time point to obtain the time contribution value; Based on the contribution value at each time point, the weighted target range time percentage is calculated, and the weighted target range time percentage is determined as the second stability index; The first stability index and the second stability index are combined to obtain the blood glucose regulation stability index.

5. The method according to claim 1, characterized in that, The process of generating stability evaluation results based on the blood glucose regulation stability index includes: Input the first and second stability indices from the blood glucose regulation stability indices into the aggregation function to generate a comprehensive stability score. Based on the mean and variance parameters of blood glucose at each moment in the blood glucose probability distribution data stream, a simulated blood glucose trajectory is generated using the Monte Carlo simulation method. Calculate the overall stability score for each simulated blood glucose trajectory, and perform statistical analysis on all the overall stability scores in conjunction with a preset statistical significance level to generate confidence intervals for each overall stability score; By combining the comprehensive stability scores and the corresponding confidence intervals, the stability evaluation results are obtained.

6. An intelligent evaluation system for intraoperative blood glucose regulation stability, characterized in that, The system includes: The data processing module is used to synchronize multimodal intraoperative data in time to obtain multidimensional synchronized time-series data, and to summarize the multidimensional synchronized time-series data to generate a multidimensional synchronized time-series dataset; wherein, the multidimensional synchronized time-series dataset includes core blood glucose measurement values, tissue perfusion and oxygenation indicators, drug infusion records and physiological rhythm indicators. The confidence index calculation module is used to calculate confidence component indices based on the multidimensional synchronous time series dataset, and combine all the confidence component indices to generate a set of confidence component indices; wherein, the confidence component indices include drug interference confidence component indices, physiological interference confidence component indices, physiological consistency confidence component indices, and reference comparison confidence component indices. The parameter generation module is used to dynamically fuse the set of confidence component indicators to generate probability distribution parameters; wherein, the probability distribution parameters are used to characterize the uncertainty of the current blood glucose measurement value; The data conversion module is used to convert the core blood glucose measurement value stream into a blood glucose probability distribution data stream based on the probability distribution parameters. The stability index generation module is used to calculate the blood glucose regulation stability index based on the blood glucose probability distribution data stream. The result generation module is used to generate a stability evaluation result based on the blood glucose regulation stability index; wherein the stability evaluation result includes a comprehensive stability score and a confidence interval of the comprehensive stability score.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images