Anesthesia hypotension prediction method and system based on grey system FMD data mining

Abstract

The application discloses an anesthesia hypotension prediction method and system based on FMD data mining of a grey system, and belongs to the field of electric digital data processing. The application carries out preprocessing, interval grey number conversion and dimension standardization processing on brachial artery blood flow mediated vasodilation function data and clinical baseline data, generates a standardized interval grey number feature set, completes feature screening and coupling relationship mining through grey nonlinear correlation analysis, completes risk simulation calculation through a non-equidistant metabolism GM(1, N) power model, and divides and outputs a hypotension risk level after anesthesia induction relying on grey clustering evaluation. The system is electrically connected by multiple functional modules in sequence, and can realize missing data completion, model precision verification and feature contribution output. The application is suitable for a poor information small sample scene, reduces the interference of data uncertainty and dimension difference, improves the reliability and interpretability of risk prediction, and provides evaluation support for perioperative hemodynamic management.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing, and in particular to a method and system for predicting hypotension under anesthesia based on FMD data mining using grey systems. Background Technology

[0002] Perioperative hemodynamic regulation is a core aspect of clinical anesthesia management. Elderly patients with hypertension exhibit more pronounced perioperative hemodynamic fluctuations due to their unique vascular physiology. Risk assessment of post-anesthesia hypotension has become a crucial research direction in perioperative anesthesia safety management. FMD, or flow-mediated dilation, is a standard abbreviation for non-invasive assessment of vascular endothelial function in anesthesiology and cardiovascular medicine. It specifically refers to multi-dimensional quantitative data collected preoperatively, including the final value of brachial artery flow-mediated dilation, peak flow velocity during reactive hyperemia, baseline vessel diameter, post-hyperemia vessel diameter, and time-series rate of diameter change. This data serves as a core input source for data mining and risk prediction analysis. As a classic indicator for non-invasive assessment of peripheral vascular endothelial function, brachial artery flow-mediated dilation has been increasingly applied in cardiovascular risk screening and clinical prognosis assessment, becoming an important data source for perioperative physiological status evaluation. With the development of medical big data and intelligent analysis technologies, technologies such as data preprocessing, feature mining, mathematical model prediction, and clustering and grading are constantly being improved. Grey system theory, with its advantages in processing data with limited information and small samples, is gradually being integrated into the field of medical risk prediction. Currently, perioperative anesthesia-related risk prediction has formed a technical framework for multi-source data integration, quantitative analysis, and risk output. Various assessment methods based on physiological data and clinical baseline data are constantly emerging, driving perioperative risk assessment towards standardization and process-orientation, and providing diversified data support for clinical anesthesia decision-making.

[0003] Existing methods for predicting post-anesthesia hypotension risk have significant limitations in data processing and prediction workflows, making them difficult to adapt to actual clinical applications. Current methods only employ conventional processing methods for brachial artery flow-mediated vasodilation data and clinical baseline data, failing to perform sequence fluctuation correction, interval gray number conversion, and dimensional standardization. This results in the inability to effectively eliminate data uncertainty and dimensional differences. Existing prediction methods largely rely on conventional feature screening methods and cannot utilize nonlinear correlation analysis to uncover feature coupling relationships, leading to insufficient extraction and utilization of feature information. Existing risk prediction models cannot adapt to non-equidistant clinical data sequences, exhibiting weak fit in risk simulation calculations. Furthermore, they lack standardized risk level classification through gray clustering assessment, resulting in a limited output format that fails to generate standardized hierarchical assessment conclusions. In information-poor scenarios, the predictive applicability and result standardization are insufficient to meet the needs of clinical anesthesia risk management. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for predicting hypotension under anesthesia based on FMD data mining of grey systems.

[0005] The objective of this invention is achieved through the following technical solution: A method for predicting hypotension during anesthesia based on FMD data mining using grey systems is provided. The method includes the following steps: S1. The brachial artery flow-mediated vasodilatory function data and clinical baseline data were subjected to sequence preprocessing, interval gray number conversion and dimensional normalization to generate a standardized interval gray number feature set. The brachial artery flow-mediated vasodilatory function data included the final value of flow-mediated vasodilation, peak blood flow velocity during reactive hyperemia, baseline vessel diameter, vessel diameter after hyperemia and time series data of diameter change rate. The clinical baseline data included age, body mass index, ASA classification, duration of hypertension treatment and hypertension classification. S2. Based on the standardized interval gray number feature set, feature selection and feature coupling relationship mining are completed through gray nonlinear correlation analysis to generate a feature set that meets the set correlation degree requirements; S3. Based on the feature set that meets the set correlation requirements, the risk simulation calculation of hypotension after anesthesia induction is completed by using the non-equidistant metabolic GM(1,N) power model, and the risk simulation value is generated. S4. Based on the risk simulation value, the risk level is classified by gray clustering assessment, and the result of the risk level of hypotension after anesthesia induction is output.

[0006] Furthermore, step S1 includes the following sub-steps: S1.1. For brachial artery blood flow-mediated vasodilation function data and clinical baseline data, sequence preprocessing is performed using an enhanced variable weighting weakening buffer operator. The weight coefficients are adaptively adjusted based on the dispersion of the data sequence. Values ​​with dispersion exceeding a set threshold are assigned adjusted weights, and the preprocessed data sequence is output. S1.2. Based on the preprocessed data sequence, and combined with the error boundary of the ultrasound measurement system and the clinically acceptable measurement deviation range, an interval gray number sequence corresponding to each value is generated; S1.3. Perform dual-dimensional normalization processing on the interval gray number sequence using gray kernel and gray level to eliminate the influence of differences in different feature dimensions and generate a standardized interval gray number feature set.

[0007] Furthermore, step S2 includes the following sub-steps: S2.1. Construct a reference sequence for the occurrence of hypotension after anesthesia induction, and calculate the gray nonlinear correlation degree between each feature sequence and the reference sequence in the standardized interval gray number feature set. The gray nonlinear correlation degree is calculated based on the second difference between the feature sequence and the reference sequence. S2.2. Based on the gray nonlinear correlation degree and a set correlation degree threshold, features whose gray nonlinear correlation degree exceeds the set threshold are selected to generate an initial feature set; S2.3. Calculate the gray feature coupling degree between every two feature sequences in the initial feature set, generate coupled feature terms that meet the set coupling degree threshold, add the coupled feature terms to the initial feature set, and generate a feature set that meets the set correlation degree requirements.

[0008] Furthermore, step S3 includes the following sub-steps: S3.1. Construct a non-equidistant metabolic GM(1,N) power model. The non-equidistant metabolic GM(1,N) power model is set up with an input layer, a background value reconstruction layer, a power index optimization layer, a parameter update layer and an output layer connected in sequence. The input layer is connected to a feature set that meets the set correlation requirements. The background value reconstruction layer reconstructs the background value of the non-equidistant sequence through the Simpson numerical integral formula, thus constructing the basic structure of the non-equidistant GM(1,N) power model. S3.2. The power exponent optimization layer uses a chaotic particle swarm optimization algorithm to traverse and optimize, and determines the nonlinear power exponent of the non-equidistant metabolic GM(1,N) power model. S3.3. The parameter update layer adopts a fixed sequence length rolling modeling method. When new sample data is added, the earliest sample data is removed through the metabolic mechanism, and the parameters of the non-equidistant metabolic GM(1,N) power model are dynamically updated. S3.4. The output layer uses a non-equidistant metabolic GM(1,N) power model to simulate and calculate the risk of hypotension after anesthesia induction, generating simulated risk values.

[0009] Furthermore, step S4 includes the following sub-steps: S4.1. Define the gray class level of the risk of hypotension after anesthesia induction and the risk clustering interval boundary corresponding to each gray class level. The gray class level includes three categories: low risk, medium risk and high risk. Each gray class level is set with an independent whitening weight function parameter. S4.2. Based on the simulated risk values, the clustering coefficients for each gray class level are calculated using the gray clustering evaluation with the center point mixed triangular whitening weight function; S4.3. Based on the clustering coefficient and the set maximum clustering coefficient judgment rule, determine the risk level of post-anesthesia hypotension corresponding to the risk simulation value, and output the post-anesthesia hypotension risk level result.

[0010] Furthermore, in step S1, for brachial artery blood flow-mediated vasomotor function data and clinical baseline data with missing data, data completion is performed using an adaptive grey Verhulst completion model. The adaptive grey Verhulst completion model is set up with a missing rate determination layer, a completion strategy matching layer, a sequence completion layer, and a result output layer connected in sequence. Based on the data missing rate and the set missing rate threshold, the corresponding completion strategy is selected. For data with a missing rate lower than the set threshold, single-sequence grey Verhulst completion is used, and for data with a missing rate higher than or equal to the set threshold, multi-feature coupled grey Verhulst completion is used. Combined with the grey nonlinear correlation between features, features that meet the set requirements are introduced as auxiliary sequences. After completion, the complete data sequence is output.

[0011] Furthermore, in step S2.1, the gray nonlinear correlation degree is calculated based on the second-order difference between the feature sequence and the reference sequence. Before the calculation, the feature sequence of the standardized interval gray number feature set and the reference sequence of the hypotension occurrence outcome after anesthesia induction are subjected to the same direction and normalization processing. The second-order difference sequence between the processed feature sequence and the reference sequence is calculated respectively. The gray nonlinear correlation degree is calculated based on the absolute value of the difference between the second-order difference sequence. The gray nonlinear correlation degree is used to characterize the degree of curve convexity similarity between the feature sequence and the reference sequence.

[0012] Furthermore, in step S3.3, after the non-equal-interval metabolic GM(1,N) power model is constructed, the prediction accuracy of the non-equal-interval metabolic GM(1,N) power model is verified by the grey posterior error test method. The posterior error ratio and small error probability of the non-equal-interval metabolic GM(1,N) power model are calculated. Based on the posterior error ratio and small error probability, the prediction accuracy level of the non-equal-interval metabolic GM(1,N) power model is divided. Non-equal-interval metabolic GM(1,N) power models that meet the set accuracy level requirements are selected. At the same time, the stability of the non-equal-interval metabolic GM(1,N) power model is verified by the Bootstrap resampling of a set number of times. The corrected prediction accuracy of the non-equal-interval metabolic GM(1,N) power model after resampling is calculated.

[0013] Furthermore, in step S4.3, while outputting the risk level result of hypotension after anesthesia induction, the contribution ratio of the gray nonlinear correlation degree of each feature to the risk level result of hypotension after anesthesia induction is also output. The contribution ratio of the gray nonlinear correlation degree is calculated based on the ratio of the gray nonlinear correlation degree of each feature to the sum of the gray nonlinear correlation degrees of all features. All features are arranged in descending order of contribution ratio, and the corresponding list of features and the corresponding contribution ratio value are output.

[0014] An anesthesia hypotension prediction system based on gray system FMD data mining is provided. The system includes a gray normalization preprocessing module, a gray nonlinear feature screening module, a gray risk simulation calculation module, a gray missing data completion module, and a result output module that are electrically connected in sequence. The gray normalization preprocessing module is used to complete data sequence preprocessing, interval gray number conversion and standardized interval gray number feature set generation; the gray nonlinear feature screening module is used to complete gray nonlinear correlation degree calculation, feature screening and feature set generation that meets the set correlation degree requirements; the gray risk simulation calculation module is used to complete the simulation calculation of hypotension risk after anesthesia induction; the gray missing data completion module is used to complete the adaptive completion of missing data; and the result output module is used to complete the risk level classification and result output.

[0015] The grey nonlinear correlation analysis involved in this scheme is a method based on grey system theory to conduct sequence nonlinear trend matching analysis. It is used to identify the implicit correlation between multi-dimensional features and target outcomes, and provides the core analytical logic for feature selection in this scheme. The grey nonlinear correlation degree is the quantitative output of the above analysis method, used to characterize the degree of matching between the changing trends of a single feature sequence and the anesthesia hypotension outcome sequence. In this scheme, it serves as the core criterion for feature retention or removal. The grey feature coupling degree is used to quantify the degree of nonlinear interaction between two feature sequences. In this scheme, it is used to explore the synergistic relationship between multiple features and generate supplementary coupling feature terms to enrich the information dimensions of the feature set. The risk clustering interval boundary is a pre-defined numerical interval endpoint corresponding to different risk levels. In this scheme, it is used to divide the numerical coverage range of risk levels and provide basic parameter support for subsequent risk level clustering determination.

[0016] The beneficial effects of this invention are: (1) By preprocessing brachial artery blood flow-mediated vasodilatory function data and clinical baseline data, gray feature mining, risk simulation calculation and gray clustering classification, the standardized classification output of hypotension risk after anesthesia induction is realized, providing a basic assessment basis for perioperative risk management. (2) Grey nonlinear correlation analysis is used to explore the implicit coupling relationship between features, and the risk fitting is completed by combining the nonlinear grey model, which weakens the interference caused by data uncertainty and improves the accuracy of risk prediction in the scenario of poor information and small sample. (3) By dynamically updating the model and outputting the feature contribution, the dynamic changes in clinical data can be adapted to make the prediction results interpretable, providing stable support for the individualized management of perioperative hemodynamics. Attached Figure Description

[0017] Figure 1A flowchart illustrating the steps of a method for predicting hypotension during anesthesia based on FMD data mining using grey systems. Figure 2 The following is a flowchart illustrating the specific steps of a method for predicting hypotension during anesthesia based on FMD data mining using a grey system, provided as an example. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 See Figure 1 This embodiment provides a method for predicting hypotension during anesthesia based on FMD data mining of grey systems. The method includes the following steps: S1. The brachial artery flow-mediated vasodilatory function data and clinical baseline data were subjected to sequence preprocessing, interval gray number conversion and dimensional normalization to generate a standardized interval gray number feature set. The brachial artery flow-mediated vasodilatory function data included the final value of flow-mediated vasodilation, peak blood flow velocity during reactive hyperemia, baseline vessel diameter, vessel diameter after hyperemia and time series data of diameter change rate. The clinical baseline data included age, body mass index, ASA classification, duration of hypertension treatment and hypertension classification. S2. Based on the standardized interval gray number feature set, feature selection and feature coupling relationship mining are completed through gray nonlinear correlation analysis to generate a feature set that meets the set correlation degree requirements; S3. Based on the feature set that meets the set correlation requirements, the risk simulation calculation of hypotension after anesthesia induction is completed by using the non-equidistant metabolic GM(1,N) power model, and the risk simulation value is generated. S4. Based on the risk simulation value, the risk level is classified by gray clustering assessment, and the result of the risk level of hypotension after anesthesia induction is output.

[0020] In some embodiments, step S1 includes the following sub-steps: S1.1. For brachial artery blood flow-mediated vasodilation function data and clinical baseline data, sequence preprocessing is performed using an enhanced variable weighting weakening buffer operator. The weight coefficients are adaptively adjusted based on the dispersion of the data sequence. Values ​​with dispersion exceeding a set threshold are assigned adjusted weights, and the preprocessed data sequence is output. S1.2. Based on the preprocessed data sequence, and combined with the error boundary of the ultrasound measurement system and the clinically acceptable measurement deviation range, an interval gray number sequence corresponding to each value is generated; S1.3. Perform dual-dimensional normalization processing on the interval gray number sequence using gray kernel and gray level to eliminate the influence of differences in different feature dimensions and generate a standardized interval gray number feature set.

[0021] In some embodiments, step S2 includes the following sub-steps: S2.1. Construct a reference sequence for the occurrence of hypotension after anesthesia induction, and calculate the gray nonlinear correlation degree between each feature sequence and the reference sequence in the standardized interval gray number feature set. The gray nonlinear correlation degree is calculated based on the second difference between the feature sequence and the reference sequence. S2.2. Based on the gray nonlinear correlation degree and a set correlation degree threshold, features whose gray nonlinear correlation degree exceeds the set threshold are selected to generate an initial feature set; S2.3. Calculate the gray feature coupling degree between every two feature sequences in the initial feature set, generate coupled feature terms that meet the set coupling degree threshold, add the coupled feature terms to the initial feature set, and generate a feature set that meets the set correlation degree requirements.

[0022] In some embodiments, step S3 includes the following sub-steps: S3.1. Construct a non-equidistant metabolic GM(1,N) power model. The non-equidistant metabolic GM(1,N) power model is set up with an input layer, a background value reconstruction layer, a power index optimization layer, a parameter update layer and an output layer connected in sequence. The input layer is connected to a feature set that meets the set correlation requirements. The background value reconstruction layer reconstructs the background value of the non-equidistant sequence through the Simpson numerical integral formula, thus constructing the basic structure of the non-equidistant GM(1,N) power model. S3.2. The power exponent optimization layer uses a chaotic particle swarm optimization algorithm to traverse and optimize, and determines the nonlinear power exponent of the non-equidistant metabolic GM(1,N) power model. S3.3. The parameter update layer adopts a fixed sequence length rolling modeling method. When new sample data is added, the earliest sample data is removed through the metabolic mechanism, and the parameters of the non-equidistant metabolic GM(1,N) power model are dynamically updated. S3.4. The output layer uses a non-equidistant metabolic GM(1,N) power model to simulate and calculate the risk of hypotension after anesthesia induction, generating simulated risk values.

[0023] In some embodiments, step S4 includes the following sub-steps: S4.1. Define the gray class level of the risk of hypotension after anesthesia induction and the risk clustering interval boundary corresponding to each gray class level. The gray class level includes three categories: low risk, medium risk and high risk. Each gray class level is set with an independent whitening weight function parameter. S4.2. Based on the simulated risk values, the clustering coefficients for each gray class level are calculated using the gray clustering evaluation with the center point mixed triangular whitening weight function; S4.3. Based on the clustering coefficient and the set maximum clustering coefficient judgment rule, determine the risk level of post-anesthesia hypotension corresponding to the risk simulation value, and output the post-anesthesia hypotension risk level result.

[0024] In some embodiments, in step S1, for brachial artery blood flow-mediated vasomotor function data and clinical baseline data with missing data, data completion is performed using an adaptive grey Verhulst completion model. The adaptive grey Verhulst completion model is configured with a missing rate determination layer, a completion strategy matching layer, a sequence completion layer, and a result output layer connected in sequence. Based on the data missing rate and a set missing rate threshold, a corresponding completion strategy is selected. For data with a missing rate lower than the set threshold, single-sequence grey Verhulst completion is used, and for data with a missing rate higher than or equal to the set threshold, multi-feature coupled grey Verhulst completion is used. Features that meet the set requirements are introduced as auxiliary sequences based on the grey nonlinear correlation between features. After completion, a complete data sequence is output.

[0025] In some embodiments, in step S2.1, the gray nonlinear correlation degree is calculated based on the second-order difference between the feature sequence and the reference sequence. Before the calculation, the feature sequence of the standardized interval gray number feature set and the reference sequence of the hypotension occurrence outcome after anesthesia induction are subjected to the same direction and normalization processing. The second-order difference sequence between the processed feature sequence and the reference sequence is calculated respectively. The gray nonlinear correlation degree is calculated based on the absolute value of the difference between the second-order difference sequence. The gray nonlinear correlation degree is used to characterize the degree of curve convexity similarity between the feature sequence and the reference sequence.

[0026] In some embodiments, in step S3.3, after the non-equal-interval metabolic GM(1,N) power model is constructed, the prediction accuracy of the non-equal-interval metabolic GM(1,N) power model is verified by the grey posterior error test method. The posterior error ratio and small error probability of the non-equal-interval metabolic GM(1,N) power model are calculated. Based on the posterior error ratio and small error probability, the prediction accuracy level of the non-equal-interval metabolic GM(1,N) power model is divided. Non-equal-interval metabolic GM(1,N) power models that meet the set accuracy level requirements are selected. At the same time, the stability of the non-equal-interval metabolic GM(1,N) power model is verified by the Bootstrap resampling of a set number of times. The corrected prediction accuracy of the non-equal-interval metabolic GM(1,N) power model after resampling is calculated.

[0027] In some embodiments, in step S4.3, while outputting the risk level result of hypotension after anesthesia induction, the contribution ratio of the gray nonlinear correlation degree of each feature to the risk level result of hypotension after anesthesia induction is also output. The contribution ratio of the gray nonlinear correlation degree is calculated based on the ratio of the gray nonlinear correlation degree of each feature to the sum of the gray nonlinear correlation degrees of all features. All features are arranged in descending order of contribution ratio, and the corresponding list of features and the corresponding contribution ratio value are output.

[0028] An anesthesia hypotension prediction system based on gray system FMD data mining is provided. The system includes a gray normalization preprocessing module, a gray nonlinear feature screening module, a gray risk simulation calculation module, a gray missing data completion module, and a result output module that are electrically connected in sequence. The gray normalization preprocessing module is used to complete data sequence preprocessing, interval gray number conversion and standardized interval gray number feature set generation; the gray nonlinear feature screening module is used to complete gray nonlinear correlation degree calculation, feature screening and feature set generation that meets the set correlation degree requirements; the gray risk simulation calculation module is used to complete the simulation calculation of hypotension risk after anesthesia induction; the gray missing data completion module is used to complete the adaptive completion of missing data; and the result output module is used to complete the risk level classification and result output.

[0029] Example 2 This embodiment provides a specific implementation process for a method to predict anesthetic hypotension based on FMD data mining using a grey system. This embodiment systematically processes preoperative brachial artery blood flow-mediated vasodilation data and clinical baseline data to complete the graded prediction of the risk of hypotension after anesthesia induction, providing corresponding data support for perioperative hemodynamic management. Figure 2 As shown, the implementation process of this embodiment includes the following steps: Step 1. Data preprocessing and standardized feature set generation: This step is used to preprocess, characterize, and normalize the original input data, eliminating the effects of abnormal data fluctuations, dimensional differences, and missing information, and generating a standardized feature set that can be used for subsequent analysis. This step includes the following sub-steps: Step 1.1. Raw Data Sequence Buffer Preprocessing: The input data in this step includes two categories. The first category is brachial artery blood flow-mediated vasodilation data, which includes the final value of blood flow-mediated vasodilation, peak blood flow velocity during reactive hyperemia, baseline vessel diameter, post-hyperemia vessel diameter, and time-series data on the rate of change of diameter. The second category is clinical baseline data, which includes age, body mass index, ASA classification, duration of hypertension treatment, and hypertension classification.

[0030] Flow-mediated vasodilation refers to the ability of vascular endothelium to release vasoactive substances under the shear force of blood flow, mediating vasodilation. It is a commonly used indicator for assessing vascular endothelial function. In this embodiment, the quantitative data corresponding to this indicator is used as one of the core input data. This step preprocesses the input raw data sequence using an enhanced variable weight weakening buffer operator. The enhanced variable weight weakening buffer operator is a preprocessing tool in grey system theory used to weaken abnormal fluctuations in data sequences and correct sequence trends. It can adaptively adjust the weights based on the degree of sequence dispersion, avoiding the over-correction of data trends by fixed weight operators. In this embodiment, it is used to perform sequence smoothing preprocessing on the raw input data.

[0031] The specific implementation process is as follows: First, the dispersion of each set of original data sequences is calculated. The dispersion is characterized by the coefficient of variation of the sequence, which is the ratio of the sequence standard deviation to the sequence mean. Based on the magnitude of the coefficient of variation, corresponding weight coefficients are adaptively generated. The larger the coefficient of variation, the smaller the weight coefficient of the corresponding value. Values ​​with dispersion exceeding the set threshold are assigned adjusted weights. The values ​​in the sequence are weighted and corrected using the weight coefficients. After the weight correction is completed for all values ​​in the sequence, the preprocessed data sequence is output.

[0032] In some embodiments, a fixed-weight weakening buffer operator can be used instead of an enhanced variable-weight weakening buffer operator to complete the data sequence preprocessing. The fixed-weight weakening buffer operator smooths the data sequence based on a preset fixed weight coefficient and is suitable for scenarios where the dispersion of the data sequence fluctuates within a small range.

[0033] In some specific implementations, to address the issue of insufficient matching between dispersion determination and weight adjustment during the buffer preprocessing of the original data sequence, the sequences of brachial artery blood flow-mediated vasomotor function data and clinical baseline data are first segmented and split. The single long sequence is split into multiple consecutive short sequences according to the data acquisition time sequence, and the coefficient of variation of each short sequence is calculated. Then, the coefficients of variation of all short sequences are integrated to obtain the dispersion characterization value of the overall sequence. Based on this characterization value, the weight coefficients of the enhanced variable weight weakening buffer operator are generated. Short sequence segments with dispersion exceeding the set threshold are independently weighted and corrected, while segments that do not exceed the set threshold retain their original numerical weights. After correction, all short sequences are spliced ​​into a complete sequence, and the preprocessed data sequence is output. This implementation method can improve the correction effect of local abnormal fluctuations in the data sequence, maintain the integrity of the overall trend of the data sequence, and avoid the loss of effective data features caused by global weight correction.

[0034] Step 1.2. Interval Gray Number Conversion of Preprocessed Data: Interval gray number is a fundamental concept in gray system theory used to describe numerical values ​​with incomplete information and a defined range. It refers to gray numbers where only the range of values ​​is known, but the exact value is unknown. In this embodiment, it is used to characterize the uncertainty of measurement data and preserve the original information boundaries of the data. This step generates an interval gray number sequence corresponding to each value based on the preprocessed data sequence and in combination with the error boundary of the ultrasound measurement system and the clinically acceptable measurement deviation range. Specifically, for each single-point value in the preprocessed data sequence, taking that value as the center, and combining the error boundary of the ultrasound measurement system and the clinically acceptable measurement deviation range, the upper and lower bounds of the interval gray number corresponding to that value are determined. After converting all single-point values ​​into the corresponding interval gray numbers, the complete interval gray number sequence is output.

[0035] Step 1.3. Normalization of the Interval Gray Number Sequence: This step performs dual-dimensional normalization of the interval gray number sequence using both gray kernel and gray level to eliminate the influence of differences in different feature dimensions, generating a standardized interval gray number feature set. The gray kernel is the mean of the upper and lower bounds of the interval gray number, used to characterize the central value of the gray number; the gray level is the ratio of the interval length of the gray number to the gray kernel, used to characterize the degree of uncertainty of the gray number. In this embodiment, dual-dimensional normalization using gray kernel and gray level eliminates the influence of different feature dimensions, ensuring that different types of feature data can be used for subsequent unified analysis.

[0036] The specific implementation process is as follows: First, the gray kernel of the interval gray number sequence is subjected to dimensionless normalization processing, and the gray kernel value is mapped to the set value interval. Then, the gray level of the interval gray number sequence is subjected to synchronous normalization processing to ensure that the normalization processing of gray kernel and gray level are consistent with the mapping rules. After all interval gray number sequences have completed the two-dimensional normalization processing, the standardized interval gray number feature set is output.

[0037] Step 1.4. Adaptive completion of missing data: This step addresses missing data in brachial artery flow-mediated vasomotor function and clinical baseline data by using an adaptive grey Verhulst completion model. The grey Verhulst model is a predictive model in grey system theory used to fit saturation-type and S-shaped change sequences. It is suitable for completing physiological data sequences with growth saturation characteristics. In this embodiment, it is used to complete missing input data, ensuring the integrity of all input data and preventing subsequent analysis from being impossible due to missing data.

[0038] The adaptive gray Verhulst completion model is set up with a missing rate determination layer, a completion strategy matching layer, a sequence completion layer, and a result output layer connected in series. The connection relationship between each layer is as follows: the output of the missing rate determination layer is connected to the input of the completion strategy matching layer, the output of the completion strategy matching layer is connected to the input of the sequence completion layer, and the output of the sequence completion layer is connected to the input of the result output layer.

[0039] The specific implementation process of this step is as follows: First, the missing data rate of the input data sequence is calculated through the missing rate determination layer. The missing data rate is the ratio of the number of missing data in the sequence to the total amount of data in the sequence. Then, the completion strategy matching layer selects the corresponding completion strategy based on the missing data rate and the set missing rate threshold. For data with a missing rate lower than the set threshold, single-sequence gray Verhulst completion is used. A gray Verhulst differential equation is constructed based on the existing complete data in the sequence. After solving the parameters of the differential equation, the missing data completion calculation is completed through the equation obtained by solving. For data with a missing rate higher than or equal to the set threshold, multi-feature coupled gray Verhulst completion is used. High correlation features that meet the set requirements are introduced as auxiliary sequences in combination with the gray nonlinear correlation between features. A multivariate gray Verhulst differential equation is constructed. After solving the equation parameters, the missing data completion calculation is completed. Then, the sequence completion layer completes the missing data completion calculation based on the selected completion strategy. Finally, the complete data sequence after completion is output through the result output layer.

[0040] In some embodiments, the traditional GM(1,1) model can be used to replace the adaptive gray Verhulst completion model to complete missing data. The traditional GM(1,1) model is suitable for data sequence completion with a monotonic trend and can simplify the calculation steps of the completion process.

[0041] Step 2. Grey Nonlinear Relational Analysis and Feature Set Construction: This step is used to mine the correlation between each feature in the standardized interval gray number feature set and the occurrence of hypotension after anesthesia induction, screen effective features, explore the coupling relationship between features, and generate a feature set that can be used for risk prediction. This step includes the following sub-steps: Step 2.1. Reference Sequence Construction and Grey Nonlinear Relationship Calculation: This step first constructs a reference sequence for the occurrence of hypotension after anesthesia induction, and then calculates the grey nonlinear relationship between each feature sequence and the reference sequence in the standardized interval grey number feature set. Grey nonlinear relationship is a branch of grey relational analysis, calculating the similarity of convexity between sequences based on the second-order difference of the sequence curves. It is used to characterize the consistency of the nonlinear change trends of two sequences. In this embodiment, it is used to quantify the correlation between the feature sequence and the outcome sequence, capturing the consistency of nonlinear change trends that traditional linear relational analysis cannot identify.

[0042] The specific implementation process is as follows: First, the feature sequences of the standardized interval gray number feature set and the reference sequence of the hypotension occurrence outcome after anesthesia induction are homogenized, converting all sequences into sequences that grow in the same direction. Then, the homogenized sequences are normalized to map the values ​​of all sequences to the interval [0,1], eliminating the influence of sequence magnitude differences on the calculation results. Next, the first-order difference sequences of the processed feature sequences and the reference sequences are calculated separately. The second-order difference sequence is calculated based on the first-order difference sequence. The gray nonlinear correlation degree is calculated based on the absolute value of the difference between the second-order difference sequences. The gray nonlinear correlation degree takes the value range of [0,1]. The closer the value is to 1, the higher the degree of similarity of the curve convexity between the feature sequence and the reference sequence. After completing the gray nonlinear correlation degree calculation for all feature sequences, the gray nonlinear correlation degree result for each feature is output.

[0043] In some embodiments, gray absolute correlation can be used instead of gray nonlinear correlation to calculate the degree of sequence correlation. Gray absolute correlation calculates the correlation degree based on the geometric similarity of the sequence curves and is suitable for sequence correlation analysis with linear trends.

[0044] In some specific implementations, to address the insufficient adaptability of sequence homogenization processing in the gray nonlinear correlation degree calculation process, the trend of the feature sequence of the standardized interval gray number feature set and the reference sequence of the hypotension occurrence outcome after anesthesia induction are first determined to identify the growth, decline, or stationary change trend of the sequence. The declining trend sequence is converted into the growth trend sequence by numerical inversion, while the stationary trend sequence retains its original numerical form. After completing the trend unification, normalization processing is performed, and the processed sequence values ​​are mapped to a set interval. Then, the first-order difference and second-order difference sequences are calculated, and the gray nonlinear correlation degree is calculated based on the absolute value of the difference between the second-order difference sequences. This implementation method can eliminate the influence of sequence trend differences on the correlation degree calculation results, improve the accuracy of the determination of the convexity similarity of the feature sequence and the reference sequence curves, and ensure the rationality of subsequent feature selection.

[0045] Step 2.2. Feature Filtering Based on Relevance Threshold: This step filters features whose gray nonlinear correlation exceeds the set threshold based on gray nonlinear correlation, generating an initial feature set. Specifically, a correlation threshold is first set. The gray nonlinear correlation corresponding to each feature is compared with the set threshold. Features whose gray nonlinear correlation exceeds the threshold are retained, while features whose gray nonlinear correlation does not reach the threshold are removed. After all eligible features have been filtered, the initial feature set is output.

[0046] Table 1. Gray Nonlinear Relationship Feature Filtering Rules In some specific implementations, the feature selection and coupling feature generation rules based on gray nonlinear correlation are shown in Table 1. Feature classification and subsequent processing are completed according to the set correlation intervals. Features with gray nonlinear correlation falling within the [0.8, 1] interval are identified as core features and retained in the initial feature set. They can be coupled with any other features that meet the set requirements to generate corresponding coupling feature terms. Features with gray nonlinear correlation falling within the [0.5, 0.8) interval are identified as secondary features and retained in the initial feature set. They can only be coupled with core features and cannot generate coupling feature terms with other secondary features or features to be selected. Features with gray nonlinear correlation falling within the [0.2, 0.5) interval are identified as features to be selected and included in the elimination feature set. They are not included in the initial feature set, cannot participate in the generation of any coupling feature terms, and are not included in the subsequent risk prediction model input. Features whose gray nonlinear correlation degree falls within the range [0, 0.2) are deemed invalid features, are all included in the feature set to be removed, are not included in the initial feature set, cannot participate in the generation of any coupled feature terms, and are not included in the subsequent risk prediction model input.

[0047] After completing the screening and classification of all features, based on the retained core and secondary features, the subsequent feature coupling degree calculation and coupling feature term generation are completed.

[0048] Step 2.3. Feature Coupling Relationship Mining and Feature Set Generation: This step calculates the gray feature coupling degree between every two feature sequences in the initial feature set, generates coupled feature terms that meet the set coupling degree threshold, and supplements the initial feature set with these coupled feature terms to generate a feature set that meets the set correlation requirements. Gray feature coupling degree is a quantitative indicator of the nonlinear coupling effect between two feature sequences based on gray nonlinear correlation degree calculation. It is used to characterize the degree of interaction between two features. In this embodiment, it is used to mine implicit coupling relationships between features, generate coupled feature terms, and enrich the information dimensions of the feature set.

[0049] The specific implementation process is as follows: First, calculate the gray feature coupling degree between every two feature sequences in the initial feature set. Then, compare the calculated gray feature coupling degree with the set coupling degree threshold to generate coupled feature items whose gray feature coupling degree exceeds the set coupling degree threshold. Add the coupled feature items to the initial feature set. After the feature set is expanded, output the feature set that meets the set correlation requirements.

[0050] Step 3. Grey Model Construction and Risk Simulation Calculation: This step involves constructing a grey prediction model adapted to the characteristics of clinical data, simulating the risk of hypotension after anesthesia induction, and generating simulated risk values ​​that can be used for risk classification. This step includes the following sub-steps: Step 3.1. Construction of the basic structure of the grey prediction model: This step constructs the non-equidistant metabolic GM(1,N) power model. The non-equidistant metabolic GM(1,N) power model is an extension of the grey prediction model. It introduces a non-linear power exponent on the basis of the traditional GM(1,N) linear model, and adapts to the needs of non-equidistant data sequences and dynamic sample updates. In this embodiment, it is used to construct the core calculation model for risk prediction.

[0051] The non-equal-interval metabolic GM(1,N) power model is configured with an input layer, a background value reconstruction layer, a power-law optimization layer, a parameter update layer, and an output layer connected in series. The connection relationships between each layer are as follows: the output of the input layer is connected to the input of the background value reconstruction layer; the output of the background value reconstruction layer is connected to the input of the power-law optimization layer; the output of the power-law optimization layer is connected to the input of the parameter update layer; and the output of the parameter update layer is connected to the input of the output layer. The specific implementation process is as follows: the input layer receives the feature set generated in step 2 that meets the set correlation requirements, and transmits the feature set data to the background value reconstruction layer. The background value reconstruction layer reconstructs the background values ​​of the non-equal-interval sequence using the Simpson numerical integral formula, thus constructing the basic structure of the non-equal-interval GM(1,N) power model.

[0052] Simpson's numerical integral formula is a numerical integration method that fits the integrand using a quadratic polynomial. Compared to the trapezoidal integral formula, it has higher computational accuracy. In this embodiment, it is used to reconstruct the background values ​​of non-equally spaced sequences, thereby improving the model's fitting accuracy.

[0053] Table 2. Hierarchical Parameters of the Non-Equally Interval Metabolic GM(1,N) Power Model Model hierarchy Input data types Core processing logic Output data type Data flow direction Input layer Multidimensional feature sequences Data format validation and normalization Standardized feature matrix Flow to background value reconstruction layer Background value reconstruction layer Standardized feature matrix Simpson Numerical Integral Background Value Reconstruction Reconstructed background value sequence Flow to the power exponent optimization layer Power-index optimization layer Reconstructed background value sequence Parameter optimization in chaotic particle swarm optimization algorithm Optimal nonlinear power exponent Flow Parameter Update Layer Parameter update layer Optimal nonlinear power exponent Rolling metabolic parameter updates Optimized model parameter set Flow to output layer Output layer Optimized model parameter set Risk value simulation calculation Anaesthesia hypotension risk simulation value Output to gray clustering stage In some specific implementations, the hierarchical settings and data flow rules of the non-equidistant metabolic GM(1,N) power model are shown in Table 2. Each level completes the sequential processing and flow of data according to the set connection relationship. The input layer receives multi-dimensional feature sequences that meet the set correlation requirements. It first performs format verification of the input data, removes abnormal data that does not meet the format requirements, and then performs secondary normalization on the verified feature sequences to generate a standardized feature matrix. The standardized feature matrix is ​​then output to the background value reconstruction layer. After receiving the standardized feature matrix, the background value reconstruction layer extracts the sampling time nodes and corresponding values ​​of the feature sequences. It then performs integral fitting on the non-equidistant sampling sequences using the Simpson numerical integral formula to reconstruct the background value sequence corresponding to the sequence. The reconstructed background value sequence is then output to the power exponent optimization layer.

[0054] After receiving the reconstructed background value sequence, the power exponent optimization layer uses a chaotic particle swarm optimization algorithm to traverse and optimize the nonlinear power exponent based on the set optimization range and objective, determining the optimal nonlinear power exponent and outputting it to the parameter update layer. The parameter update layer, receiving the optimal nonlinear power exponent, dynamically updates the model parameters based on a fixed-length rolling training sequence through a metabolic mechanism, generating an optimized model parameter set, which is then output to the output layer. The output layer, receiving the optimized model parameter set, simulates the risk of hypotension after anesthesia induction, generating the corresponding simulated risk value, which is then output to the subsequent grey clustering evaluation stage.

[0055] Step 3.2. Determination of the Nonlinear Power Exponent of the Model: This step uses a chaotic particle swarm optimization algorithm to traverse and optimize, determining the nonlinear power exponent of the non-equidistant metabolic GM(1,N) power model. The chaotic particle swarm optimization algorithm is an optimization algorithm that combines the ergodicity of chaotic motion with the fast optimization capability of particle swarm optimization. It is used for global optimization of nonlinear parameters, avoiding the problem of traditional optimization algorithms getting stuck in local optima. In this embodiment, it is used to determine the nonlinear power exponent of the model, breaking the linear assumption of the traditional GM model and adapting to the nonlinear relationship between features and risks. Specifically, the optimization range of the nonlinear power exponent is first set. The chaotic particle swarm optimization algorithm is then used to traverse and optimize within this range, with the minimum model fitting residual as the optimization objective, to determine the optimal nonlinear power exponent of the non-equidistant metabolic GM(1,N) power model. After determining the power exponent, the parameters are transferred to the parameter update layer.

[0056] Step 3.3. Dynamic update of model parameters and model validation: This step adopts a rolling modeling method with a fixed sequence length. When new sample data is added, the oldest sample data is removed through the metabolic mechanism, and the parameters of the non-equidistant metabolic GM(1,N) power model are dynamically updated.

[0057] The specific implementation process is as follows: First, the length of the training sequence is set. When new sample data is added, the new sample data is added to the training sequence, while the oldest sample data in the training sequence is removed to keep the length of the training sequence fixed. The parameters of the model are recalculated based on the updated training sequence to complete the dynamic update of the parameters. After the non-equidistant metabolic GM(1,N) power model is constructed, the prediction accuracy of the non-equidistant metabolic GM(1,N) power model is verified by the grey posterior error test method. The posterior error ratio and small error probability of the non-equidistant metabolic GM(1,N) power model are calculated. Based on the posterior error ratio and small error probability, the prediction accuracy level of the non-equidistant metabolic GM(1,N) power model is divided. Non-equidistant metabolic GM(1,N) power models that meet the set accuracy level requirements are selected. At the same time, the stability of the non-equidistant metabolic GM(1,N) power model is verified by the Bootstrap resampling of a set number of times. The corrected prediction accuracy of the non-equidistant metabolic GM(1,N) power model after resampling is calculated.

[0058] In some embodiments, cross-validation can be used instead of Bootstrap resampling to complete model stability verification. Cross-validation divides the sample set into multiple subsets and uses the subsets as validation sets in turn to complete model verification, which is suitable for scenarios with relatively sufficient sample size.

[0059] In some specific implementations, to address the issue of insufficient matching degree of the rolling sequence during the dynamic update of parameters in the non-equidistant metabolic GM(1,N) power model, a fixed length of the model training rolling sequence is first set. The feature dimensions and numerical formats of the newly added sample data are extracted. The format of the newly added sample data is aligned and the dimensions are unified with the historical sample data in the training sequence. Sample data with mismatched formats are removed. Then, the newly added sample data that meets the requirements is added to the end of the training sequence. Simultaneously, the sample data with the earliest start time in the training sequence is removed to keep the length of the training sequence constant. Based on the updated training sequence, the parameters of the non-equidistant metabolic GM(1,N) power model are refitted. After the parameter update is completed, the model parameters are transmitted to the output layer. This implementation method can ensure the consistency of the model training sequence format and the validity of the data, improve the smoothness of the dynamic update of model parameters, and maintain the stability of the model calculation process.

[0060] Step 3.4. Simulation Calculation of Post-anesthesia Hypotension Risk: This step uses a non-equidistant metabolic GM(1,N) power model to simulate the risk of post-anesthesia hypotension and generate simulated risk values. In this step, the input to the non-equidistant metabolic GM(1,N) power model is a feature set that meets the set correlation requirements, and the output is the simulated risk value corresponding to post-anesthesia hypotension. The algorithm and business logic are combined by fitting the non-linear relationship between feature data and the risk of post-anesthesia hypotension using the non-equidistant metabolic GM(1,N) power model, converting multi-dimensional feature data into single-dimensional simulated risk values ​​that can be used for risk level classification, providing basic data for subsequent risk level classification. Specifically, the feature set data to be predicted is input into the constructed non-equidistant metabolic GM(1,N) power model. The model completes the risk simulation through hierarchical calculations and finally outputs the corresponding simulated risk value through the output layer.

[0061] Step 4. Grey Clustering Assessment and Risk Result Output: This step performs grey clustering assessment on the simulated risk values, classifies the risk levels, and outputs the final risk level results and corresponding feature contribution information. This step includes the following sub-steps: Step 4.1. Setting of risk gray class levels and corresponding parameters: This step sets the gray class level of hypotension risk after anesthesia induction and the risk clustering interval boundary corresponding to each gray class level. The gray class level includes three categories: low risk, medium risk and high risk. Each gray class level is set with an independent whitening weight function parameter.

[0062] Gray class levels are categories used in gray clustering assessment to classify the levels of research subjects. In this embodiment, they are used to classify the risk levels of post-anesthesia induction hypotension, achieving a graded risk output. The specific implementation process is as follows: First, three gray class levels for post-anesthesia induction hypotension risk are set: low risk, medium risk, and high risk. Then, a corresponding risk clustering interval boundary is set for each gray class level, and the numerical range corresponding to each gray class level is determined. Finally, an independent whitening weight function parameter is set for each gray class level. After the parameter setting is completed, the parameters are transferred to the subsequent clustering calculation stage.

[0063] Table 3. Gray Classification Parameters for the Risk of Hypotension After Anesthesia Induction Gray class levels Risk clustering interval boundary interval Whitening weight function center point Left turn Right turn Low risk [0,0.3] 0.15 0 0.3 Medium risk [0.2,0.7] 0.45 0.2 0.7 High risk [0.6,1] 0.8 0.6 1 Invalid value range (-∞,0)∪(1,+∞) - - - In some specific implementations, the gray class classification and whitening weight function parameter settings for the risk of hypotension after anesthesia induction are shown in Table 3. Based on the set risk clustering interval boundary interval and whitening weight function parameters, the center point mixed triangular whitening weight function corresponding to each gray class level is constructed. For the low-risk gray class level, with the set center point as the benchmark, the left turning point coincides with the left endpoint of the risk clustering interval boundary interval, and the right turning point coincides with the right endpoint of the risk clustering interval boundary interval, thus constructing the corresponding triangular whitening weight function. When the risk simulation value falls within the low-risk risk clustering interval boundary interval, the output value of the whitening weight function decreases linearly as the distance between the risk simulation value and the center point increases.

[0064] For the medium-risk gray category, using a set center point as a reference, the left and right turning points correspond to the left and right endpoints of the risk clustering interval boundary, respectively. The constructed triangular whitening weight function reaches its maximum value at the center point and decreases linearly to the left and right, with overlapping intervals with the whitening weight functions of the low-risk and high-risk gray categories, used to cover the fuzzy transition area of ​​the risk boundary. For the high-risk gray category, using a set center point as a reference, the left and right turning points correspond to the left and right endpoints of the risk clustering interval boundary, respectively. The constructed triangular whitening weight function shows a trend of first increasing and then decreasing as the risk simulation value increases. For risk simulation values ​​within the invalid value interval, the whitening weight function output value is 0 and does not participate in the subsequent clustering coefficient calculation and risk level determination.

[0065] Step 4.2. Grey Clustering Calculation Based on Whitening Weight Function: This step calculates the clustering coefficient for each grey class level corresponding to the simulated risk value using a centroid-mixed triangular whitening weight function for grey clustering evaluation. The centroid-mixed triangular whitening weight function is a function used in grey clustering evaluation to map observed values ​​to different grey classes. A triangular membership function is constructed based on the centroid of the grey class, used to calculate the clustering coefficient for each grey class corresponding to the observed value. In this embodiment, it is used to calculate the clustering coefficient for each grey class level corresponding to the simulated risk value, providing a basis for subsequent risk level determination.

[0066] The specific implementation process is as follows: Based on the gray class levels, risk clustering interval boundaries, and whitening weight function parameters set in step 4.1, a centroid mixed triangular whitening weight function is constructed for each gray class level. The simulated risk values ​​are substituted into the whitening weight function for each gray class level to calculate the clustering coefficient for each gray class level. After calculating all clustering coefficients, the clustering coefficient results for each gray class level are output. In some embodiments, a weighted gray clustering evaluation method can be used instead of the centroid mixed triangular whitening weight function gray clustering evaluation. The weighted gray clustering evaluation method completes the clustering calculation based on preset feature weights and is suitable for scenarios where the importance of features has a clear clinical consensus.

[0067] Step 4.3. Risk Level Determination and Result Output: This step determines the risk level of post-anesthesia hypotension corresponding to the simulated risk value based on the clustering coefficient and the set maximum clustering coefficient judgment rule, and outputs the post-anesthesia hypotension risk level result.

[0068] The specific implementation process is as follows: The clustering coefficients of each gray class level corresponding to the simulated risk value are compared. Based on the maximum clustering coefficient determination rule, the gray class level corresponding to the maximum clustering coefficient is determined as the post-anesthesia induction hypotension risk level corresponding to that simulated risk value, and the post-anesthesia induction hypotension risk level result is output. Simultaneously with outputting the post-anesthesia induction hypotension risk level result, the contribution ratio of each feature to the gray nonlinear correlation degree of the post-anesthesia induction hypotension risk level result is also output. The contribution ratio of gray nonlinear correlation degree is calculated based on the ratio of the gray nonlinear correlation degree of each feature to the sum of the gray nonlinear correlation degrees of all features. All features are arranged in descending order of contribution ratio, and the corresponding arranged feature list and corresponding contribution ratio values ​​are output.

[0069] This embodiment constructs a prediction process for post-anesthesia induction hypotension risk based on grey system theory. Through grey preprocessing of input data, nonlinear correlation mining of features, and grey prediction and grading of risk, it achieves risk prediction in scenarios with limited information and small sample sizes. This embodiment preserves the original uncertainty of measurement data through interval grey number transformation, reducing the interference of measurement errors on prediction results. Grey nonlinear correlation analysis uncovers the nonlinear correlation between features and outcomes, as well as the coupling relationship between features, improving the utilization efficiency of feature information. A nonlinear simulation calculation of risk is achieved through a non-equidistant metabolic GM(1,N) power model, which is adaptable to non-equidistant clinical data sequences and supports dynamic updates of model parameters, allowing model iteration without full sample retraining. Grey clustering assessment achieves risk grading and outputs the feature contribution ratio, improving the interpretability of prediction results. The overall process of this embodiment can be implemented without a large number of labeled samples, adapting to the risk prediction needs of different clinical scenarios and providing stable data support for perioperative hemodynamic management.

[0070] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for predicting hypotension during anesthesia based on FMD data mining using grey systems, characterized in that, Includes the following steps: S1. The brachial artery flow-mediated vasodilatory function data and clinical baseline data were subjected to sequence preprocessing, interval gray number conversion and dimensional normalization to generate a standardized interval gray number feature set. The brachial artery flow-mediated vasodilatory function data included the final value of flow-mediated vasodilation, peak blood flow velocity during reactive hyperemia, baseline vessel diameter, vessel diameter after hyperemia and time series data of diameter change rate. The clinical baseline data included age, body mass index, ASA classification, duration of hypertension treatment and hypertension classification. S2. Based on the standardized interval gray number feature set, feature selection and feature coupling relationship mining are completed through gray nonlinear correlation analysis to generate a feature set that meets the set correlation degree requirements; S3. Based on the feature set that meets the set correlation requirements, the risk simulation calculation of hypotension after anesthesia induction is completed by using the non-equidistant metabolic GM(1,N) power model, and the risk simulation value is generated. S4. Based on the risk simulation value, the risk level is classified by gray clustering assessment, and the result of the risk level of hypotension after anesthesia induction is output.

2. The method according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1.

1. For brachial artery blood flow-mediated vasodilation function data and clinical baseline data, sequence preprocessing is performed using an enhanced variable weighting weakening buffer operator. The weight coefficients are adaptively adjusted based on the dispersion of the data sequence. Values ​​with dispersion exceeding a set threshold are assigned adjusted weights, and the preprocessed data sequence is output. S1.

2. Based on the preprocessed data sequence, and combined with the error boundary of the ultrasound measurement system and the clinically acceptable measurement deviation range, an interval gray number sequence corresponding to each value is generated; S1.

3. Perform dual-dimensional normalization processing on the interval gray number sequence using gray kernel and gray level to eliminate the influence of differences in different feature dimensions and generate a standardized interval gray number feature set.

3. The method according to claim 1, characterized in that, Step S2 includes the following sub-steps: S2.

1. Construct a reference sequence for the occurrence of hypotension after anesthesia induction, and calculate the gray nonlinear correlation degree between each feature sequence and the reference sequence in the standardized interval gray number feature set. The gray nonlinear correlation degree is calculated based on the second difference between the feature sequence and the reference sequence. S2.

2. Based on the gray nonlinear correlation degree and a set correlation degree threshold, features whose gray nonlinear correlation degree exceeds the set threshold are selected to generate an initial feature set; S2.

3. Calculate the gray feature coupling degree between every two feature sequences in the initial feature set, generate coupled feature terms that meet the set coupling degree threshold, add the coupled feature terms to the initial feature set, and generate a feature set that meets the set correlation degree requirements.

4. The method according to claim 1, characterized in that, Step S3 includes the following sub-steps: S3.

1. Construct a non-equidistant metabolic GM(1,N) power model. The non-equidistant metabolic GM(1,N) power model is set up with an input layer, a background value reconstruction layer, a power index optimization layer, a parameter update layer and an output layer connected in sequence. The input layer is connected to a feature set that meets the set correlation requirements. The background value reconstruction layer reconstructs the background value of the non-equidistant sequence through the Simpson numerical integral formula, thus constructing the basic structure of the non-equidistant GM(1,N) power model. S3.

2. The power exponent optimization layer uses a chaotic particle swarm optimization algorithm to traverse and optimize, and determines the nonlinear power exponent of the non-equidistant metabolic GM(1,N) power model. S3.

3. The parameter update layer adopts a fixed sequence length rolling modeling method. When new sample data is added, the earliest sample data is removed through the metabolic mechanism, and the parameters of the non-equidistant metabolic GM(1,N) power model are dynamically updated. S3.

4. The output layer uses a non-equidistant metabolic GM(1,N) power model to simulate and calculate the risk of hypotension after anesthesia induction, generating simulated risk values.

5. The method according to claim 1, characterized in that, Step S4 includes the following sub-steps: S4.

1. Define the gray class level of the risk of hypotension after anesthesia induction and the risk clustering interval boundary corresponding to each gray class level. The gray class level includes three categories: low risk, medium risk and high risk. Each gray class level is set with an independent whitening weight function parameter. S4.

2. Based on the simulated risk values, the clustering coefficients for each gray class level are calculated using the gray clustering evaluation with the center point mixed triangular whitening weight function; S4.

3. Based on the clustering coefficient and the set maximum clustering coefficient judgment rule, determine the risk level of post-anesthesia hypotension corresponding to the risk simulation value, and output the post-anesthesia hypotension risk level result.

6. The method according to claim 2, characterized in that, In step S1, for brachial artery blood flow-mediated vasomotor function data and clinical baseline data with missing data, data completion is performed using an adaptive grey Verhulst completion model. The adaptive grey Verhulst completion model is set up with a missing rate determination layer, a completion strategy matching layer, a sequence completion layer, and a result output layer connected in sequence. Based on the data missing rate and the set missing rate threshold, the corresponding completion strategy is selected. For data with a missing rate lower than the set threshold, single-sequence grey Verhulst completion is used, and for data with a missing rate higher than or equal to the set threshold, multi-feature coupled grey Verhulst completion is used. Features that meet the set requirements are introduced as auxiliary sequences based on the grey nonlinear correlation between features. After completion, the complete data sequence is output.

7. The method according to claim 3, characterized in that, In step S2.1, the gray nonlinear correlation degree is calculated based on the second-order difference between the feature sequence and the reference sequence. Before the calculation, the feature sequence of the standardized interval gray number feature set and the reference sequence of the hypotension occurrence outcome after anesthesia induction are subjected to the same direction and normalization processing. The second-order difference sequence of the processed feature sequence and the reference sequence are calculated respectively. The gray nonlinear correlation degree is calculated based on the absolute value of the difference between the second-order difference sequence. The gray nonlinear correlation degree is used to characterize the degree of curve convexity similarity between the feature sequence and the reference sequence.

8. The method according to claim 4, characterized in that, In step S3.3, after the non-equal-interval metabolic GM(1,N) power model is constructed, the prediction accuracy of the non-equal-interval metabolic GM(1,N) power model is verified by the grey posterior error test method. The posterior error ratio and small error probability of the non-equal-interval metabolic GM(1,N) power model are calculated. Based on the posterior error ratio and small error probability, the prediction accuracy level of the non-equal-interval metabolic GM(1,N) power model is divided. Non-equal-interval metabolic GM(1,N) power models that meet the set accuracy level requirements are selected. At the same time, the stability of the non-equal-interval metabolic GM(1,N) power model is verified by Bootstrap resampling a set number of times. The corrected prediction accuracy of the non-equal-interval metabolic GM(1,N) power model after resampling is calculated.

9. The method according to claim 5, characterized in that, In step S4.3, while outputting the risk level of hypotension after anesthesia induction, the contribution ratio of the gray nonlinear correlation degree of each feature to the risk level of hypotension after anesthesia induction is also output. The contribution ratio of the gray nonlinear correlation degree is calculated based on the ratio of the gray nonlinear correlation degree of each feature to the sum of the gray nonlinear correlation degrees of all features. All features are arranged in descending order of contribution ratio, and the corresponding list of features and the corresponding contribution ratio value are output.

10. An anesthesia hypotension prediction system based on gray system-based FMD data mining, used to perform the method as described in any one of claims 1-9, characterized in that, It includes a gray normalization preprocessing module, a gray nonlinear feature screening module, a gray risk simulation calculation module, a gray missing data completion module, and a result output module, which are connected in sequence by electricity. The gray normalization preprocessing module is used to complete data sequence preprocessing, interval gray number conversion and standardized interval gray number feature set generation; the gray nonlinear feature screening module is used to complete gray nonlinear correlation degree calculation, feature screening and feature set generation that meets the set correlation degree requirements; the gray risk simulation calculation module is used to complete the simulation calculation of hypotension risk after anesthesia induction; the gray missing data completion module is used to complete the adaptive completion of missing data; and the result output module is used to complete the risk level classification and result output.

Images