Preoperative risk assessment method for department of cardiology

Abstract

The invention relates to the technical field of physiological signal prediction, in particular to a preoperative risk assessment method for the department of cardiology, which comprises the following steps: sliding window segmentation time sequence data to calculate a baseline offset, dynamic time warping alignment parameter fluctuation rate to generate an abnormal mark, and standard deviation comparison amplitude threshold triggering risk signals. K-means clustering multi-dimensional data mapping risk levels, isolated forest detection abnormal fluctuation and electrocardiogram and myocardial zymogram cross validation are combined, and a preoperative risk assessment conclusion is output. According to the method, individual differences and interferences are eliminated through a sliding window algorithm, a time difference problem is solved by aligning a multi-parameter fluctuation rate through dynamic time warping, a quantitative evaluation standard is established by comparing a standard deviation with an amplitude threshold value to avoid limitation of a single threshold value, and objective grading is realized by combining K-means clustering with an Euclidean distance. The isolated forest algorithm, the electrocardiogram ST segment and myocardial zymogram cross validation form a multi-modal evaluation system, and the risk evaluation sensitivity and specificity are improved in a complete closed-loop mode.

Description

Technical Field

[0001] The present invention relates to the technical field of physiological signal prediction, and in particular to a method for preoperative risk assessment in cardiology. Background Art

[0002] The field of physiological signal prediction technology encompasses methods for monitoring and predicting human health using a variety of physiological parameters and signals. Physiological signals, such as electrocardiograms, blood pressure, pulse, and respiratory rate, are collected through sensors or monitoring equipment. The core of physiological signal prediction technology lies in real-time analysis and processing of the collected data to extract health information that can serve as a warning. This technology is primarily used in medical monitoring, disease prevention, and health management, providing support for medical decision-making through scientific analysis and prediction of potential health risks. Key aspects of physiological signal prediction technology include signal acquisition, signal processing, data analysis, and model prediction. Its goal is to provide the conditions for early intervention through accurate data prediction.

[0003] Among them, the cardiology preoperative risk assessment method refers to a technical means of evaluating a patient's preoperative risk by analyzing their physiological signals. This topic mainly addresses technical issues in preoperative risk assessment, specifically covering the use of patients' physiological signal data, such as electrocardiograms, blood pressure, pulse, etc., combined with clinical background information, to conduct a multi-dimensional analysis of their preoperative status. This method establishes a mathematical model, combines the patient's original health data with real-time monitoring data, and analyzes and predicts the patient's health risks before surgery. In this way, it can provide medical personnel with an objective basis for preoperative risk assessment and enhance the scientific nature of surgical safety management.

[0004] Existing technologies for analyzing physiological signals often rely on static thresholds based on a single parameter and lack a dynamic multi-parameter coordination mechanism. This makes it difficult to effectively capture abnormal correlations between differentiated physiological parameters. For example, fluctuations in electrocardiogram (ECG) and blood pressure data vary in time, and traditional methods lack time alignment, making them prone to misjudgment. Existing risk assessment models often use fixed or empirical thresholds, which are unable to adapt to individual physiological differences and are prone to missed detections or false alarms when patients' baseline levels fluctuate significantly. Risk grading often relies on expert experience or simple segmentation rules, lacking a data-driven quantitative classification basis. This results in overly general grading results that are difficult to support refined clinical decision-making. Existing methods for verifying abnormal signals primarily rely on single-modality data and fail to fully integrate clinical indicators such as the ECG ST segment and myocardial enzyme profile. This creates data silos and reduces the clinical interpretability of prediction results. For example, when an abnormal myocardial enzyme profile is not correlated with real-time physiological signals, early detection of myocardial ischemia is delayed. These shortcomings lead to problems such as an imbalance in sensitivity and specificity and delayed risk assessment in practical applications of existing technologies. Summary of the Invention

[0005] In order to solve the problem that the existing technology for analyzing physiological signals mostly relies on single-parameter static threshold judgment and lacks a dynamic coordination mechanism for multiple parameters, which makes it difficult to effectively capture the abnormal correlation between differentiated physiological parameters. For example, the fluctuations of electrocardiogram and blood pressure data have time dimension differences. Traditional methods do not perform time alignment processing, which is prone to misjudgment. Existing risk assessment models often use fixed thresholds or empirical thresholds, which cannot adapt to the differences in individual physiological characteristics and are prone to missed detection or false alarms when the patient's baseline level fluctuates greatly. Risk level classification mostly relies on expert experience or simple segmentation rules and lacks data-driven quantitative classification basis, resulting in overly general classification results and difficulty in supporting refined clinical decision-making. Existing methods mainly rely on single-modality data for the verification of abnormal signals and do not fully integrate clinical indicators such as electrocardiogram ST segment and myocardial enzyme spectrum. There is a data island phenomenon, which reduces the clinical interpretability of the prediction results. For example, when the myocardial enzyme spectrum is abnormal but not correlated with real-time physiological signals, the early judgment of myocardial ischemia is delayed. The above shortcomings lead to technical problems such as imbalance between sensitivity and specificity and lag in risk assessment in practical applications of existing technologies. The embodiment of the present invention provides a preoperative risk assessment method for cardiology. The technical solution is as follows:

[0006] In one aspect, a method for preoperative risk assessment in cardiology is provided, the method comprising:

[0007] S1: Obtain the patient's preoperative heart rate, blood pressure, and respiratory rate physiological parameters through medical monitoring equipment. Use a sliding window algorithm to segment the time series data, calculate the median within the window as the baseline, and offset the physiological parameter data from the baseline to obtain the fluctuation offset.

[0008] S2: Based on the fluctuation offset, perform time alignment processing based on the dynamic time warping algorithm, calculate the fluctuation rate of each parameter, determine whether there is a trend abnormal change, and obtain a trend abnormality mark;

[0009] S3: Calling the trend anomaly marker, extracting the corresponding fluctuation offset to evaluate the fluctuation amplitude, and comparing it with the amplitude threshold. When the fluctuation amplitudes of two or more items exceed the amplitude threshold, a risk trigger signal is generated;

[0010] S4: Call the multidimensional standard deviation data in the risk trigger signal, use the K-means clustering algorithm for classification, set three cluster centers, use the Euclidean distance as the standard to calculate the cluster center value size mapping risk level, and output the risk level label;

[0011] S5: Call the risk level label, detect the abnormal degree of its fluctuation offset through the isolation forest algorithm, combine the electrocardiogram ST segment and myocardial enzyme spectrum detection data to perform clinical indication cross-validation, and output the preoperative risk assessment conclusion.

[0012] As a further solution of the present invention, the sliding window duration of the sliding window algorithm is 24 hours before surgery, and the window sliding step length is 5 minutes;

[0013] In the K-means clustering algorithm, the input data needs to be Z-score standardized before clustering, and the standard deviation screening threshold is set based on actual data and clinical statistics to output low, medium, and high risk level labels;

[0014] The trend anomaly is based on a fluctuation rate threshold set at 2 times the standard deviation of the original data mean. When the rates of two or more parameters exceed the threshold, it is determined to be a trend anomaly.

[0015] The isolation forest algorithm combines the fluctuation trajectory obtained by the dynamic time warping algorithm with the abnormal period to perform spatiotemporal analysis;

[0016] The fluctuation offset specifically refers to the heart rate baseline offset, blood pressure baseline offset, and respiratory rate baseline offset. The trend abnormality mark includes a unidirectional continuous trend, a bidirectional fluctuation trend, and an intermittent abnormal trend. The risk trigger signal includes a dual-parameter collaborative exceedance signal, a three-parameter joint exceedance signal, and a continuous window trigger signal. The risk level label specifically refers to a low-risk threshold interval, a medium-risk transition interval, and a high-risk warning interval. The risk assessment conclusion includes a cardiac function reserve index, a respiratory failure risk value, and a multiple organ dysfunction score.

[0017] As a further solution of the present invention, the specific steps of S1 include:

[0018] S101: Acquire physiological data recorded by monitoring equipment, including heart rate, blood pressure, and respiratory rate, sort the data based on timestamps, segment the time series data using a sliding window, calculate the median within the window, and generate a window benchmark median sequence;

[0019] S102: Based on the window reference median sequence, call the time series data, perform offset calculation point by point, obtain the offset between the data and the window reference value, and generate a physiological parameter offset sequence;

[0020] S103: extracting a large offset value, a small offset value, and a mean offset value from the offset data of the heart rate, blood pressure, and respiratory rate in the physiological parameter offset sequence, marking the corresponding monitoring times in chronological order, and calculating the offset fluctuation range of each physiological parameter to obtain a fluctuation offset;

[0021] The fluctuation range of the offset is compared with the original data 72 hours before the operation for comparative analysis to obtain the baseline reference value of the movement line.

[0022] As a further solution of the present invention, the offset is calculated using the formula:

[0023]

[0024] Among them, Δ n Represents the offset of the nth time point, x n Represents the physiological parameter measurement value at the nth time point, M w Represents the median value of the wth window in the sliding window benchmark median sequence, μ w represents the arithmetic mean of the physiological parameter measurements within the wth sliding window, σ w represents the physiological parameter measurement value in the wth sliding window, and β is a dynamic adjustment factor determined based on the patient's age and gender, with a value range of 0.8-1.2.

[0025] As a further solution of the present invention, the specific steps of S2 include:

[0026] S201: Based on the fluctuation offset, a dynamic time warping algorithm is used to perform time series alignment processing on the two-period parameter values, identify the synchronous offset trajectory of the parameters at different time points, and generate an aligned time series feature value;

[0027] S202: Calculating the change amplitude of the parameter in a continuous time period based on the aligned time series feature value, extracting the change slope of the parameter value in the time period, and comparing the change slope of each parameter with its baseline slope in the aligned time series to calculate the slope deviation of the parameter in the time period to obtain a slope offset difference value;

[0028] The baseline slope is obtained by training the parameter change rate under normal conditions 72 hours before surgery;

[0029] S203: Based on the slope offset difference value, a trend change identification threshold is set, and whether the parameter difference value exceeds the trend change identification threshold is determined. The parameters in the time period that continuously exceed the threshold are uniformly marked and classified to generate a trend abnormality mark;

[0030] The trend change identification threshold is set with reference to clinical standards and based on expert experience.

[0031] As a further solution of the present invention, the specific steps of S3 include:

[0032] S301: Calling the abnormal fluctuation mark data set, extracting the fluctuation offset value of the parameter within the marked time period, obtaining the offset sequence of the parameter at the time node, fitting the offset vector to calculate the offset variance, judging the fluctuation stability based on the offset variance, and obtaining the fluctuation offset difference data;

[0033] S302: Calculate the standard deviation of the parameter offset value within the marked time period based on the fluctuation offset difference data, call the amplitude threshold set by the parameter, calculate the parameter whose standard deviation value exceeds the corresponding amplitude threshold, and obtain offset amplitude threshold deviation data;

[0034] S303: Calling the deviation amplitude threshold value exceeding tolerance data, setting a lower limit for determining the number of exceeding tolerances, comparing the number of marked parameters within the time period according to the lower limit, and generating a risk trigger signal based on a joint determination of the parameter deviation amplitude and the number of exceeding tolerances;

[0035] The amplitude threshold is set to twice the standard deviation of the offset within the original sliding window of the parameter.

[0036] As a further solution of the present invention, the specific steps of S4 include:

[0037] S401: Invoke the risk trigger signal, obtain multidimensional standard deviation data for a required time period, set three cluster centers based on clustering requirements, calculate the Euclidean distance value from each data point to the three cluster centers, classify the data points into corresponding categories using the shortest distance method, and obtain the Euclidean distance mapping result;

[0038] S402: Based on the Euclidean distance mapping result, extract the distance values ​​corresponding to the data points within the cluster category, use the maximum and minimum distance differences to divide the risk gradient within and between groups, set the distance range boundaries, and classify the three cluster center values ​​into three risk levels respectively, to obtain risk level classification data;

[0039] S403: According to the risk level classification data, mapping and matching are performed on the corresponding relationship between each cluster label and the risk level of the cluster center, and the label classified by each data point is replaced accordingly to obtain a risk level label.

[0040] As a further solution of the present invention, the Euclidean distance value is calculated using the formula:

[0041]

[0042] Among them, d ij Represents the Euclidean distance value from the i-th data point to the j-th cluster center, which is a dimensionless parameter, x ik Represents the standard deviation data value of the i-th data point in the k-th dimension, which is a dimensionless parameter after min-max normalization. jk Represents the coordinate value of the j-th cluster center in the k-th dimension, which is a dimensionless parameter, σ k represents the standard deviation of the k-th dimension data, which is a dimensionless parameter after min-max standardization processing, m represents the total number of data dimensions, which is a dimensionless parameter, and α represents the dynamic adjustment coefficient, which is a dimensionless parameter.

[0043] As a further solution of the present invention, the specific steps of S5 include:

[0044] S501: calling the risk level label, detecting the degree of abnormal deviation fluctuation based on the isolation forest algorithm, extracting the fluctuation deviation data of the abnormal parameters, filtering the abnormal fluctuations based on the detection results and marking the parameters, and obtaining the abnormal deviation fluctuation data;

[0045] S502: Based on the abnormal deviation fluctuation data, according to the electrocardiogram ST segment and myocardial enzyme spectrum detection data, extract the physiological indication data of each abnormal parameter, calculate the cross-validation index according to the detection data, mark the cross-validation parameters that meet the abnormal conditions, and obtain abnormal cross-validation data;

[0046] The electrocardiogram ST segment and myocardial enzyme spectrum detection data were Z-score standardized and unified;

[0047] S503: Based on the abnormal cross-validation data, a preoperative risk analysis is performed on the cross-validation parameters, and a preoperative risk assessment conclusion is obtained by combining the common characteristics of fluctuation deviation and abnormal physiological indicators.

[0048] As a further solution of the present invention, the cross-validation index is calculated using the formula:

[0049]

[0050] Among them, V q Represents the cross-validation data of the qth parameter, Zs q Represents the Z-score normalized value of the electrocardiogram ST segment detection data corresponding to the qth parameter, Zm q represents the Z-score normalized value of the myocardial enzyme spectrum detection data corresponding to the qth parameter, ΔA q Represents the abnormal value of the offset fluctuation of the qth parameter, μ A Represents the arithmetic mean of the parameter deviation fluctuation anomaly, Zs q,k Represents the normalized value of the ST segment of the electrocardiogram at the kth time node, Zm q,k represents the standardized value of the myocardial enzyme spectrum of the qth parameter at the kth time node, and n represents the total number of time nodes.

[0051] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0052] A sliding window algorithm is used to segment time series data and calculate baseline offsets, effectively eliminating the impact of individual differences and transient interference on physiological parameters and enhancing data stability. A dynamic time warping algorithm is used to align multi-parameter fluctuation rates, addressing the misjudgment of abnormalities caused by inconsistent temporal dimensions in traditional methods and improving the accuracy of multi-parameter collaborative analysis. A quantitative assessment standard for multi-dimensional fluctuation amplitudes is established based on a comparison mechanism between standard deviation and amplitude thresholds, avoiding the limitations of single threshold judgment and enhancing the robustness of anomaly detection. Multi-dimensional standard deviation data is classified using a K-means clustering algorithm, and risk levels are mapped using Euclidean distance, achieving data-driven objective grading and reducing reliance on subjective experience. The isolation forest algorithm is combined with the detection of abnormal fluctuation offsets and cross-validated with electrocardiogram ST segment and myocardial enzyme spectrum data, breaking through the bottleneck of traditional single-dimensional analysis to form an assessment system that integrates multimodal data and enhances the clinical applicability of preoperative risk prediction. These technical approaches form a complete closed loop from data preprocessing, dynamic alignment, quantitative assessment, to multimodal validation, improving the sensitivity and specificity of risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0054] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0055] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0056] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0057] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0058] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0059] See also Figure 1 The embodiment of the present invention provides a method for preoperative risk assessment in cardiology. The processing flow of the method may include the following steps:

[0060] S1: Obtain the patient's preoperative heart rate, blood pressure, and respiratory rate physiological parameters through medical monitoring equipment. Use a sliding window algorithm to segment the time series data, calculate the median within the window as the baseline, and offset the physiological parameter data from the baseline to obtain the fluctuation offset.

[0061] S2: Based on the fluctuation offset, time alignment is performed using the dynamic time warping algorithm to calculate the fluctuation rate of each parameter, determine whether there is a trend abnormal change, and obtain a trend abnormality mark;

[0062] S3: Call the trend anomaly mark, extract the corresponding fluctuation offset to evaluate the fluctuation amplitude, and compare it with the amplitude threshold. When the fluctuation amplitudes of two or more items exceed the amplitude threshold, a risk trigger signal is generated.

[0063] S4: Call the multidimensional standard deviation data in the risk trigger signal, use the K-means clustering algorithm for classification, set three cluster centers, use the Euclidean distance as the standard to calculate the cluster center value size mapping risk level, and output the risk level label;

[0064] S5: Call the risk level label, detect the abnormal degree of its fluctuation offset through the isolation forest algorithm, combine the electrocardiogram ST segment and myocardial enzyme spectrum test data to perform clinical indication cross-validation, and output the preoperative risk assessment conclusion;

[0065] The fluctuation offset specifically refers to the heart rate baseline offset, blood pressure baseline offset, and respiratory rate baseline offset. The trend abnormality marks include unidirectional continuous trend, bidirectional fluctuation trend, and intermittent abnormal trend. The risk trigger signals include dual-parameter collaborative exceedance signal, three-parameter joint exceedance signal, and continuous window trigger signal. The risk level label specifically refers to the low-risk threshold interval, medium-risk transition interval, and high-risk warning interval. The risk assessment conclusions include cardiac function reserve index, respiratory failure risk value, and multiple organ dysfunction score.

[0066] Specifically, the steps of S1 are:

[0067] S101: Acquire physiological data recorded by monitoring equipment, including heart rate, blood pressure, and respiratory rate, sort the data based on timestamps, segment the time series data using a sliding window, calculate the median within the window, and generate a window benchmark median sequence;

[0068] To obtain the physiological data recorded by the monitoring device, we first need to clarify the parameter types recorded by the monitoring device, including heart rate (unit: beats / minute), blood pressure composed of systolic and diastolic pressure (unit: mmHg), and respiratory rate (unit: beats / minute). The monitoring data are recorded in the form of timestamps and stored in the local database or server system. Taking a certain ECG monitoring device as an example, the heart rates recorded between 9:00 and 10:00 on May 18, 2025 are 78, 82, 85, 83, 88, 90, 92, 85, 80, and 76 beats / minute, respectively. The corresponding timestamps are recorded at 60-second intervals. The program reads the recorded data for each minute in this time period and sorts them in chronological order. The sorting process is to rearrange the data items by index. After the sorting is completed, the timestamps and their corresponding heart rate, blood pressure, and respiratory rate are matched one by one to form an ordered triple array. The sliding window technology is then used to segment the time series data. Taking a window length of 5 minutes and a sliding step of 1 minute as an example, the heart rate values ​​in the first window are 78, 82, 85, 83, and 88. The slicing logic in Python is called to remove the oldest data and add new records after each sliding. For example, the second window is 82, 85, 83, 88, and 90. Multiple sliding windows are formed by analogy. The median of the heart rate, blood pressure, and respiratory rate data in each window is calculated in turn. Taking heart rate as an example, the median of the window sequence {78, 82, 85, 83, 88} is 85, which is recorded as the benchmark value of the first window. The median calculation is to sort the values ​​in the window from small to large and take the middle value. If the number of values ​​is even, the average of the two middle values ​​is taken. For example, the sorting of the window {85, 83, 88, 90, 92} is {83, 85, 88, 90, 92}, and the median is 88. The medians of the windows are arranged in sequence to form a window benchmark median sequence, forming the following example:

[0069] Table 1 Monitoring data and median sequence table

[0070]

[0071]

[0072] As shown in Table 1, the median calculation within the sliding window can be used to smooth the fluctuations of physiological signals and serve as the benchmark value for subsequent offset calculations. Once the window median sequence is formed, it can be further used for data offset processing. The obtained median value will be used as a fixed benchmark to participate in the difference calculation with the original data point.

[0073] In the process of refining the specific execution actions of the phrases "sort", "split", and "calculate", the first sorting operation is to read the timestamp field of the recorded data and perform an ascending operation. The corresponding index array is reorganized in chronological order, so that the subsequent sliding window operation can continuously extract data along the timeline. The second splitting operation is to traverse the entire array, set the initial pointer to point to the first data point, define the window size w = 5, and the step size s = 1. Each split extracts data from the current position i to i + w - 1 and stores it as a list. The loop slides backward s units to generate the next window, and the operation is repeated until the end of the data. Finally, the median calculation operation is to take all the data in the window and call the sorting function to obtain the data item corresponding to the index value of the middle position. If the amount of data is even, the median is obtained by averaging the two middle values. Taking the heart rate window {78, 82, 85, 83, 88} as an example, the sorted result is {78, 82, 83, 85, 88}, and the median is 83, which is assigned as the window baseline value. The process of obtaining the word "median" can be further quantified: For example, for the five recorded values ​​{82, 85, 87, 90, 92}, after sorting, they become {82, 85, 87, 90, 92}, with the middle item being the third item, 87. For the six recorded values ​​{82, 85, 87, 90, 92, 95}, the sorting becomes {82, 85, 87, 90, 92, 95}, with the median being (87 + 90) / 2 = 88.5, which is then assigned to the window. During the segmentation and median calculation process, if a window contains an extreme outlier (e.g., a sudden increase of 40 beats / minute in heart rate), the original value is retained and the subsequent offset assessment step is performed.

[0074] S102: Based on the window reference median sequence, call the time series data, perform offset calculation point by point, obtain the offset between the data and the window reference value, and generate a physiological parameter offset sequence;

[0075] To perform offset processing based on the window benchmark median sequence, we first need to traverse the original record value x at each time point. n , and identify the window w to which it belongs, and extract the median M corresponding to the window w , mean μ w and standard deviation σ w , calculate the offset using the formula:

[0076]

[0077] Among them, Δ n Represents the offset of the nth time point, x n Represents the physiological parameter measurement value at the nth time point, M w Represents the median value of the wth window in the sliding window benchmark median sequence, μ w represents the arithmetic mean of the physiological parameter measurements within the wth sliding window, σw represents the physiological parameter measurement value in the wth sliding window, and β is a dynamic adjustment factor determined based on the patient's age and gender, with a value range of 0.8-1.2.

[0078] For each x n The offset calculation is performed as follows: Taking the heart rate as an example, the recorded value at 09:00 is 78 beats / minute, the window is 09:00-09:04, and the median of the window is M w =85, mean μ w =(78+82+85+83+88) / 5=83.2, standard deviation σ w The calculation is as follows:

[0079]

[0080] Assuming the patient's gender is male and age is 55 years old, and after consulting the pathological data, setting β = 1.1, we can get:

[0081]

[0082] 2.339)=1.1·2.421≈2.663;

[0083] Similarly, perform the above operations for each time point and obtain the offset to ensure that the processing logic is executed for heart rate, blood pressure, and respiratory rate respectively. Calculate the offset of systolic and diastolic pressure separately. For example, if the systolic pressure at 09:00 is 120 mmHg, the corresponding window sequence is {120, 125, 130, 128, 122}, and calculate the median M. w =125, mean μ w =125, standard deviation σ w The calculation is as follows:

[0084]

[0085] The offset is calculated as follows:

[0086]

[0087] The offset is used to characterize the dynamic fluctuation characteristics of multiple physiological parameters. After obtaining the offset data, it needs to be structured and organized according to the timestamp. For example, at 09:00, the heart rate offset is 2.663, the systolic blood pressure offset is 2.634, the diastolic blood pressure offset is 1.918, and the respiratory rate offset is 0.789. The above data is stored in the form of a vector [2.663, 2.634, 1.918, 0.789], corresponding to the time point 09:00, forming a physiological parameter offset sequence; further, each offset value is classified and divided, and the judgment interval is set according to the absolute value of the offset: if the heart rate offset value |Δ i|≤2 is defined as a weak offset, if 2<|Δ i |≤5 is a moderate deviation, and if it is greater than 5, it is a significant deviation. The interval setting here refers to the clinical statistical results of a large-sample heart rate fluctuation study. The corresponding blood pressure parameter setting interval is a slight deviation: |Δ i |≤3, moderate deviation: (3, 8], obvious deviation: greater than 8 times / minute, respiratory rate is weak deviation: |Δ i |≤2, moderate offset: (2, 5], significant offset: greater than 5 times / minute, this is used to classify offset data and generate a classifiable vector dataset; it should be noted that when the recording frequency is increased to sampling every 30 seconds, but the window is still 5 minutes and the step size is 1 minute, the number of data points in each window increases, and it is still necessary to locate the median, mean, and standard deviation of each data point in the most recent time window to which it belongs, recalculate the corresponding offset, and the recording time point accuracy is improved to the second level, which facilitates high-precision monitoring and processing of individual physiological data. The output offset value sequence should be a vector set with timestamps, for example: 09:00→[2.663, 2.634, 1.918, 0.789], 09:01→[1.322, 2.541, 1.110, 0.552], realizing time-located offset calculation and processing of all physiological parameters.

[0088] S103: extracting the large offset value, small offset value, and mean offset value from the offset data of the heart rate, blood pressure, and respiratory rate in the physiological parameter offset sequence, marking the corresponding monitoring times in chronological order, and calculating the offset fluctuation range of each physiological parameter to obtain the fluctuation offset;

[0089] The fluctuation range of the offset was compared and analyzed with the original data 72 hours before surgery to obtain the baseline reference value of the movement line;

[0090] For the offset sequence generated for each physiological parameter, its maximum value, minimum value and mean value need to be extracted item by item, and the corresponding time points need to be marked. The processing flow is to slice the vector data according to the parameter dimension. For example, for heart rate, the heart rate offset items at the time point are extracted to form an offset value array {Δ1, Δ2,…, Δ x}, execute the maximum function to select the maximum offset value max(Δ n ), call the minimum function to obtain the minimum offset value min(Δ n ), and then trace back to the time point of its occurrence through the corresponding index, and then execute the mean function: The overall trend mean is obtained. For example, the heart rate deviation sequence is {-3.2, -1.1, +0.9, +2.3, -2.5}, the maximum value is +2.3, the minimum value is -3.2, the corresponding time is 09:03 and 09:00, and the average value is μ = -3.2-1.1+0.9+2.3-2.5 / 5 = -0.72. The results show that the overall deviation of the heart rate tends to be negative. The volatility deviation is calculated as follows:

[0091] Δ range =max(Δ n )-min(Δ n )=2.3-(-3.2)=5.5,

[0092] In this way, systolic blood pressure, diastolic blood pressure and respiratory rate are processed separately, and the three types of offset extreme values ​​and means are extracted respectively. After the fluctuation offset is formed, the maximum, minimum and mean positions and corresponding values ​​of the time points are recorded through structured vectors and table structures; after obtaining the parameter fluctuation offset of the current monitoring period, it is necessary to call the database to obtain the patient's complete monitoring record for 72 hours before surgery, match the data collected by the same monitoring equipment, use the sliding window length and sliding step size that are exactly the same as the current data for segmentation and median extraction processing, reconstruct the preoperative baseline median sequence, and execute the same calculation process of S102 to obtain the preoperative baseline median sequence. The physiological parameter offset value sequence is used to extract the preoperative fluctuation offset to form the preoperative baseline data. The current monitoring fluctuation value and the preoperative fluctuation baseline value are directly subtracted. For example, if the current respiratory rate fluctuation is 7 times / minute and the preoperative baseline fluctuation is 3 times / minute, the difference is 7-3=4. The dynamic baseline reference value assigned to the parameter is +4, indicating that its fluctuation degree is higher than that before the operation. The difference value can be organized into a vector [+4, +2.5, -1.8, 0], which corresponds to physiological indicators such as heart rate, systolic blood pressure, diastolic blood pressure, and respiratory rate, respectively, for subsequent patient dynamic trend modeling and analysis.

[0093] Specifically, the steps of S2 are:

[0094] S201: Based on the fluctuation offset, a dynamic time warping algorithm is used to perform time series alignment on the parameter values ​​of the two periods, identify the synchronous offset trajectory of the parameters at different time points, and generate the aligned time series feature value;

[0095] Based on the fluctuation offset between the current cycle and the preoperative cycle, the parameter time series alignment operation needs to be performed. During the execution process, the time series sampling value of a monitoring parameter (such as heart rate) in the current cycle is first set as the main sequence. For example, the heart rate values ​​of five time points collected from 09:00 to 09:04 are 78, 81, 85, 79, and 82 beats / minute, respectively. At the same time, the same parameter sequence collected within 72 hours before the operation is selected as the control data, such as the preoperative sequence of 75, 76, 77, 76, and 78 beats / minute; then the two groups of sequences are sorted into time-parameter value key-value pairs in chronological order to ensure the consistency of time points and to ensure equal sampling intervals with minutes as the granularity. If the number of sampling points in a sequence is insufficient, it is padded by linear interpolation. In this example, the two sequences have the same length and the padded operation is skipped; then, in order to achieve accurate matching of the values, the two sequences need to be compared for positional deviation, and the distance measurement method is set to absolute difference, that is, at any time point t i Next, the current value x i Compared with the control value y i The offset is |x i -y i |, thus constructing a fifth-order distance matrix, traversing the matching path, and selecting the best mapping group based on the principle of minimum cumulative path distance. In this example, since the time points are completely aligned, a one-to-one correspondence is directly formed, and there is no need to rearrange the path. The complete alignment sequence mapping pair is formed as follows: the current 09:00 corresponds to 75 before surgery, 09:01 corresponds to 76, 09:02 corresponds to 77, 09:03 corresponds to 76, and 09:04 corresponds to 78. The parameter data in the two time periods are synchronously sorted and combined through the alignment structure to obtain the aligned time series table shown in Table 2 for subsequent processing.

[0096] S202: Calculate the change amplitude of the parameter in the continuous time period based on the aligned time series feature value, extract the change slope of the parameter value in the time period, compare the change slope of each parameter with its baseline slope in the aligned time series, calculate the slope deviation of the parameter in the time period, and obtain the slope offset difference value;

[0097] The baseline slope was obtained by training the parameter change rate under normal conditions 72 hours before surgery;

[0098] After completing the parameter alignment operation, it is necessary to further evaluate the change slope of the parameters between consecutive time points to reveal whether there is any abnormal deviation in its change trend. During the execution process, the numerical difference between two adjacent time points in the current sequence is first extracted in sequence. For example, the heart rate increases from 78 to 81 between the time points 09:00–09:01, and its change slope is 81-78=3. Similarly, 09:01–09:02 is 85-81=4, 09:02–09:03 is 79-85=-6, and 09:03–09:04 is 82-79=3. The slope of the current sequence is [3, 4, -6, 3]. The same calculation is performed on the preoperative sequence, and its slope sequence is [76-75=1, 77-76=1, 76-77=-1, 78-76=2], which is [1, 1, - 1, 2], then perform the difference operation of the two groups of slope values, and obtain the offset slope sequence as [3-1=2, 4-1=3, -6-(-1)=-5, 3-2=1], construct a complete offset difference array [2, 3, -5, 1], and then set the threshold judgment standard. Here, referring to actual monitoring experience and expert settings, the trend identification slope threshold is set to 2 times / minute, and the absolute value of each difference is processed and compared with the threshold. The judgment is made in sequence: at 09:01, the slope offset is 2, which is equal to the threshold and is not judged as exceeding; at 09:02, the offset is 3, which exceeds the threshold; at 09:03, the offset is -5, the absolute value is 5, which exceeds the threshold; at 09:04, the offset is 1, which does not exceed the threshold. A set of trend offset Boolean sequences is obtained as [no, yes, yes, no]. The specific results are shown in Table 2.

[0099] S203: Based on the slope offset difference value, a trend change identification threshold is set to determine whether the parameter difference value exceeds the trend change identification threshold, and the parameters that continuously exceed the threshold within the time period are uniformly marked and classified to generate a trend anomaly mark;

[0100] The threshold for identifying trend changes was set based on clinical standards and expert experience;

[0101] After obtaining the trend offset Boolean judgment result sequence, it is necessary to analyze whether there are continuous "yes" states in the sequence, so as to identify the time period with a continuous offset trend. The specific execution method is to traverse the slope offset judgment Boolean sequence. In the current sequence [no, yes, yes, no], it is identified that the two consecutive time points 09:02 and 09:03 are "yes". Based on this, it is confirmed that there is a trend offset behavior in the interval, and a trend anomaly mark is required. The abnormal identification process is as follows: First, record the start and end time points of the continuous "yes" interval, which is 09:02 to 09:03 here, and then The original slope data and the baseline slope data within the time period are then extracted, i.e., the current change slopes are 4 and -6, the baseline slopes are 1 and -1, and the slope offset values ​​are 3 and -5. The time period and the corresponding parameter values ​​are then entered into a structured data table and marked as "trend anomaly" for use by the subsequent analysis system. In actual operation, the process requires simultaneous analysis in combination with multi-dimensional parameters. After the marking is completed, the results are linked to the original parameter sequence and updated. The trend anomaly set can be expressed as: T = {09:02, 09:03}. This group of time periods constitutes the only trend offset interval in this embodiment.

[0102] Table 2 Alignment sequence and trend shift data

[0103]

[0104] As shown in Table 2, through trend offset identification and slope difference calculation, it is clear that there is a trend anomaly from 09:02 to 09:03 in the current period.

[0105] Specifically, the steps of S3 are:

[0106] S301: Call the abnormal fluctuation mark data set, extract the fluctuation offset value of the parameter within the marked time period, obtain the offset sequence of the parameter at the time node, fit the offset vector to calculate the offset variance, judge the fluctuation stability based on the offset variance, and obtain the fluctuation offset difference data;

[0107] After calling the trend anomaly marking dataset, it is necessary to perform offset analysis on the heart rate parameters in the marked time period from 09:02 to 09:03. First, extract the heart rate values ​​of the time period from the current cycle and the benchmark cycle respectively. The current cycle is 85 and 79, and the benchmark cycle is 77 and 76. The offset of the time point is calculated accordingly, that is, the offset of 09:02 is 85-77=8, and the offset of 09:03 is 79-76=3, forming an offset sequence [8, 3]. Then, the offset vector fitting is required for the offset sequence. The specific operation is: use the offset data as the sample vector for centering, that is, subtract the mean 8+3 / 2=5.5 respectively to obtain the centering vector [2.5, -2.5]. Then perform a square operation on the vector to obtain [6.25, 6.25], and take the average to obtain the offset variance. The calculation process =6.25+6.25 / 2=6.25, so the offset variance of this time period is 6.25. If this value is used as the fluctuation offset difference data, it needs to be compared with the difference mean in the original sample data. For example, if the original mean variance is 3.0, the difference value of this time period is significantly higher than the mean. According to the difference amplitude, it is determined that the parameter fluctuation amplitude in the current time period does not belong to the normal range. This result can be further used as the offset amplitude evaluation input for the next step of analysis. The specific offset extraction operation needs to be traversed in sequence according to the trend abnormality time period. If there are multiple abnormal segments, an independent offset sequence is constructed for each segment and its variance value is calculated. For example, the heart rate values ​​in the newly added segment 09:10 to 09:11 are 90 and 85, the corresponding benchmarks are 82 and 81, the offsets are 8 and 4, the mean is 6, and the offset variance is calculated as (8-6). 2 +(4-6) 2 / 2=4+4 / 2=4, from which it can be obtained that different segments have different degrees of offset fluctuations. The difference between the above offset variance and the mean of the entire sample is used as a measure of the degree of fluctuation offset difference.

[0108] S302: Calculate the standard deviation of the parameter's offset value within the marked time period based on the fluctuation offset difference data, call the amplitude threshold set by the parameter, calculate the parameter whose standard deviation value exceeds the corresponding amplitude threshold, and obtain offset amplitude threshold deviation data;

[0109] Based on the offset variance obtained in the previous stage as the fluctuation offset difference data, the standard deviation of the offset sequence in each abnormal time period is calculated. Specifically, the offset sequence [8, 3] from 09:02 to 09:03 is processed. The mean is first calculated to be 5.5, and the standard deviation is calculated again:

[0110]

[0111] The standard deviation is then compared with the preset offset amplitude threshold, which is defined as twice the standard deviation of the offset in the original sliding window. If the total offsets obtained in the 5-minute sliding window in the current monitoring period are [3, 4, 2, 6, 5], the standard deviation is calculated as The mean is 4 and the variance is:

[0112] The offset amplitude threshold is 2×1.41≈2.82. Since the standard deviation from 09:02 to 09:03 is 2.5, which is lower than 2.82, it is not considered as out of tolerance. If the offset of another segment is [10, 4], the mean is 7 and the standard deviation is If the value is greater than the threshold of 2.82, it is marked as out-of-tolerance data. Similarly, the standard deviation of each abnormal trend segment can be compared, and the judgment results can be recorded to generate offset amplitude threshold out-of-tolerance data. The recording format can be stored in a tag array such as [No, Yes, No], where the second segment is out of tolerance and the first and third segments are normal.

[0113] S303: Call the deviation amplitude threshold value out-of-tolerance data, set the lower limit value for the number of out-of-tolerance determination, compare the number of marked parameters in the time period based on the lower limit value, and generate a risk trigger signal based on the combined determination of the parameter deviation amplitude and the number of out-of-tolerances;

[0114] The amplitude threshold is set to 2 times the standard deviation of the offset within the original sliding window of the parameter;

[0115] After calling the deviation amplitude threshold deviation data, it is necessary to further set the lower limit value for the deviation number and make a joint judgment. In the current example, there are three abnormal trend time periods, and the lower limit value is set to 1. That is, if more than one parameter deviation occurs in the same abnormal trend segment, it is determined that the risk reaches the trigger standard. In actual operation, it is necessary to perform statistics on each abnormal trend segment. For example, the deviation standard deviation corresponding to the segment 09:02 to 09:03 is 2.5, which does not exceed the 2.82 threshold and is marked as "no". The deviation standard deviation of the other segment 09:10 to 09:11 is 3, which is marked as "yes". After calculation, it is found that there is one deviation in this segment, which is equal to the set lower limit value of 1, meeting the trigger condition and generating a risk trigger signal. If the offset of the third segment data is [5, 5] and the standard deviation is 0, it does not constitute a deviation and is not recorded as a trigger item. The entire judgment operation needs to traverse the trend abnormality segment, calculate the standard deviation and perform deviation statistics on the parameters in each segment, and compare it with the lower limit value after the statistics are completed. For example, if the lower limit value is set to 1, if the number of deviations is ≥1, a "risk trigger" signal is generated. This signal can be identified by a Boolean value and used as an input parameter for subsequent responses to form a unified risk response control mechanism.

[0116] Table 3 Statistics of trend abnormal segment offset

[0117]

[0118] As shown in Table 3, some abnormal trend segments have exceeded the threshold limit in amplitude standard deviation, are marked as out-of-tolerance and trigger corresponding risk signals.

[0119] Specifically, the steps of S4 are:

[0120] S401: Invoke the risk trigger signal to obtain multidimensional standard deviation data for the required time period. Based on clustering requirements, set three cluster centers. Calculate the Euclidean distance between each data point and the three cluster centers. Use the shortest distance method to classify the data points into corresponding categories and obtain the Euclidean distance mapping result.

[0121] The risk trigger signal is called. If an alarm message indicating abnormal preoperative signs is received, such as a rapid change in the patient's heart rate or arrhythmia within 24 hours before the operation, the system will automatically call the patient's 24-hour preoperative monitoring data, covering multiple physiological parameter dimensions such as heart rate, blood pressure, respiratory rate, and blood oxygen saturation. Each dimension obtains 24 sets of continuous data in an hourly sampling manner to form a 24×m standard deviation data matrix. Assuming m=4, it means there are four different types of preoperative physiological signal parameters. First, the 24-hour data sequence of each dimension is extracted and its standard deviation value is calculated. For example, the data sequence extracted for the heart rate dimension is [76, 79 , 82, ..., 73], the heart rate standard deviation is calculated as σ = 3.58, and the heart rate standard deviation array is σ = (3.58, 5.12, 2.77, 1.93). According to the cluster analysis requirements, the system sets three types of centers to represent low-risk preoperative status, medium-risk preoperative status, and high-risk preoperative status, respectively. According to the preoperative average eigenvalues ​​of the previous sample medical records, three groups of cluster center values ​​are constructed, set as c1 = (78, 120, 18, 98), c2 = (92, 135, 22, 95), and c3 = (105, 150, 28, 90). Each group of center points represents the average standard model of a type of preoperative status. For any preoperative data point x i , such as the monitoring standard deviation data of a patient at the 10th hour is x 10 =(81, 118, 20, 97), then the Euclidean distance between it and the three cluster centers needs to be calculated using the formula:

[0122]

[0123] Among them, d ij Represents the Euclidean distance value from the i-th data point to the j-th cluster center, which is a dimensionless parameter, x ik Represents the standard deviation data value of the i-th data point in the k-th dimension, which is a dimensionless parameter after min-max normalization. jkRepresents the coordinate value of the j-th cluster center in the k-th dimension, which is a dimensionless parameter, σ k represents the standard deviation of the k-th dimension data, which is a dimensionless parameter after min-max standardization processing, m represents the total number of data dimensions, which is a dimensionless parameter, and α represents the dynamic adjustment coefficient, which is a dimensionless parameter.

[0124] During the execution process, the difference is first extracted and standardized according to the dimension, for example Continuing to process the dimensions in this way, the normalized differences are: (0.838, 0.391, 0.722, 0.518). Squaring the four terms and adding them together gives ∑ = 0.702 + 0.153 + 0.521 + 0.268 = 1.644, and taking the square root of it gives Introducing the dynamic adjustment factor α = 0.4 and the number of dimensions m = 4, calculate the adjustment coefficient The distance to the cluster center c1 is d i1 =1.1×1.282=1.410, and so on, calculate x 10 The distances to c2 and c3 are 1.924 and 3.366, respectively. Category 1, corresponding to the minimum value of 1.410, is used as the cluster label for this data point. To ensure completeness, the system performs the above calculation process on all 24-hour data points in sequence and records the category to which the shortest distance for each data point belongs as shown below.

[0125] Table 4 Euclidean distance mapping table

[0126]

[0127] As shown in Table 4, the cluster distance results of three monitoring data points within a certain period of time before cardiology patients are listed. The category corresponding to the minimum value is the Euclidean distance mapping classification result of the point, see Table 4. This result serves as the basic data for subsequent risk level classification.

[0128] S402: Based on the Euclidean distance mapping results, the distance values ​​corresponding to the data points within the cluster category are extracted, and the maximum and minimum distance differences are used to divide the risk gradient within and between groups. The distance range boundaries are set to classify the three cluster center values ​​into three risk levels respectively, and the risk level classification data is obtained;

[0129] Based on the completed Euclidean distance mapping results, the system needs to extract the minimum Euclidean distance value of the data points in each cluster category. For example, in the preoperative evaluation scenario, category 1 contains data point x 10 、x 13 etc., and the corresponding minimum distance values ​​are 1.410, 1.295, etc., which constitute the distance value set of this category {d 10,1 ,d13,1 ,…}, in this way, the minimum Euclidean distance value sets in category 2 and category 3 are extracted respectively. The system then calculates the difference between the maximum and minimum values ​​in the minimum distance set of each type of cluster data to form the minimum-maximum distance difference of the category. For example, the maximum value in category 1 is 1.689 and the minimum value is 1.295, then the distance difference of the category is Δ1=0.394. Similarly, the distance difference between category 2 and category 3 is calculated to obtain Δ2=0.611 and Δ3=0.880. The three categories of distance differences are sorted in ascending order to obtain Δ1<Δ2< Δ3, which can be used to represent the risk gradient of the data within the group from low to high. The system then introduces a unified risk level division rule between groups, performs linear normalization on the minimum distance value interval in the category, and divides the minimum-maximum value interval of each category into three risk level intervals. For example, the distance value interval of category 1 [1.295, 1.689] is divided into: low risk [1.295, 1.426], medium risk [1.426, 1.557], and high risk [1.557, 1.689]. Based on this process, the interval boundaries are calculated using the formula:

[0130]

[0131] Among them, b i Represents the upper boundary value of the i-th partition interval, min(d j ) represents the minimum value of the Euclidean distance value in the j-th cluster category, max(d j ) represents the maximum value of the Euclidean distance value in the j-th cluster category, i represents the serial number of the division interval, and two dividing lines are calculated in the order of 1 to 2, so as to divide the entire numerical interval into three level intervals with equal distance;

[0132] Taking category 2 as an example, if its minimum value is 1.392 and its maximum value is 2.003, then the boundary

[0133] The intervals are [1.392, 1.595), [1.595, 1.797), and [1.797, 2.003], which are marked as low, medium, and high risk levels, respectively. The distance value of each data point is judged accordingly to determine its risk level label in the cluster category to which it belongs, thereby completing the risk level division of the three types of center points and the data they contain. This division process does not involve cross-class comparisons, but only relies on the risk gradient interval defined by the distance distribution characteristics within the group to achieve risk level classification data output of preoperative data.

[0134] S403: Based on the risk level classification data, mapping and matching are performed between the corresponding relationship between each cluster label and the risk level of the cluster center, and the label classified by each data point is replaced accordingly to obtain the risk level label;

[0135] According to the risk level classification data completed above, the data points corresponding to each cluster label are replaced with labels. During the execution process, a matching mapping table between cluster categories and their risk level labels needs to be established. For example, the three risk levels of category 1 are L1-1 (low), L1-2 (medium), and L1-3 (high), which correspond to the division intervals within them according to the minimum Euclidean distance. The original label of each data point is "category 1", but if its distance value in category 1 falls into the L1-2 interval, its label needs to be updated to "medium risk". Similarly, for category 2 and category 3 data, they are mapped to the three sub-levels L2-1L2-3 and L3-1L3-3 of "medium risk" and "high risk" respectively. The original category label is replaced by the above mapping to form the risk level label corresponding to each data point, such as data point x 10 The original category is 1, the distance value is 1.410, which falls into the interval [1.295, 1.426), corresponding to the low risk level, and the label is "low risk", while the data point x 11 It originally belonged to category 2, with a distance value of 1.392, falling into the interval [1.392, 1.595), and was also mapped to "low risk". It should be noted here that even if the risk level label is the same, the category it belongs to is different. The label update is only reflected in the update of the risk level and does not affect the subsequent grouping and differentiation. After the system completes the risk label update operation of the data point, the risk level label set of the preoperative data is obtained, which is used for the subsequent risk intervention strategy formulation and preoperative decision-making reference.

[0136] Specifically, the steps of S5 are:

[0137] S501: Call the risk level label, detect the degree of deviation fluctuation anomaly based on the isolation forest algorithm, extract the fluctuation deviation data of the abnormal parameters, filter the fluctuation anomaly based on the detection results and mark the parameters to obtain the deviation fluctuation anomaly data;

[0138] After calling the risk level label, it is necessary to locate the data points marked as medium and high risk in each cluster category one by one. The system first extracts the sampling data of each parameter in the data point in multiple detection cycles, and extracts the time series recording data of monitoring parameters such as heart rate, blood pressure, and blood oxygen saturation in turn, and then performs outlier detection processing between samples for each parameter. During the execution process, the system first calls the observation value set of the parameter in different detection cycles. For example, the heart rate recording value of a high-risk data point in the five cycles from T1 to T5 is [78, 80, 76, 122, 79]. The system then performs maximum, minimum and median value extraction operations on the array, and obtains 122, 76 and 79 respectively. Then the system calculates the offset value set of the parameter in the sample, that is, the absolute value of the difference between each value and the median value [1 , 1, 3, 43, 0], judge whether the offset value exceeds the preset offset reference value. For example, if the offset reference value is set to 20, the system will only retain the record with an offset value of 43 based on the judgment, and mark the corresponding period T4 as an abnormal fluctuation point. Then continue to perform the same process on the parameters, form a corresponding mark for the abnormal period corresponding to each abnormal parameter, and record the absolute value of the offset of each parameter and the positive and negative marks of its offset direction as the data structure of the extraction result. The system further accumulates and counts the number of parameters detected as abnormal in each data point, and sets the threshold of the number of abnormal parameters to 2. When two or more parameters of a data point are marked as abnormal, the system assigns an offset fluctuation abnormal label to the data point, and outputs a set of data points that meet the conditions, which is the offset fluctuation abnormality data set;

[0139] S502: Based on the deviation fluctuation abnormality data and the electrocardiogram ST segment and myocardial enzyme spectrum test data, extract the physiological indication data of each abnormal parameter, calculate the cross-validation index based on the test data, mark the cross-validation parameters that meet the abnormal conditions, and obtain abnormal cross-validation data;

[0140] The electrocardiogram ST segment and myocardial enzyme spectrum test data were standardized and quantified using the Z-score;

[0141] According to the deviation fluctuation abnormality data and the standardized electrocardiogram ST segment and myocardial enzyme spectrum detection data, the detection value of each abnormal parameter q at the specified time point needs to be extracted item by item during the execution process and Z-score normalization is performed to obtain Zs q With Zm q For example, a parameter such as "troponin I" is normalized to obtain Zs q =1.85, Zm q =0.92, deviation fluctuation abnormality ΔA q =0.58, the average value of the abnormal parameter deviation fluctuation μ A =0.64, calculate the cross validation index using the formula:

[0142]

[0143] Among them, V q Represents the cross-validation data of the qth parameter, Zs q Represents the Z-score normalized value of the electrocardiogram ST segment detection data corresponding to the qth parameter, Zm q represents the Z-score normalized value of the myocardial enzyme spectrum detection data corresponding to the qth parameter, ΔA q Represents the abnormal value of the offset fluctuation of the qth parameter, μ A Represents the arithmetic mean of the parameter deviation fluctuation anomaly, Zs q,k Represents the normalized value of the ST segment of the electrocardiogram at the kth time node, Zm q,k represents the standardized value of the myocardial enzyme spectrum of the qth parameter at the kth time node, and n represents the total number of time nodes.

[0144] The first part is calculated as Then extract the detection value of the parameter at time point k, and calculate the normalized sum of squares of the sequence differences. Assuming n = 4 time points, the detection data is as follows:

[0145] Table 5 Standardized values ​​of electrocardiogram and myocardial enzyme spectrum data

[0146] Time point ST segment normalized value Enzyme spectrum normalization value T1 1.80 1.00 T2 1.95 1.05 T3 1.60 0.98 T4 1.70 1.12

[0147] As shown in Table 5, the ST segment and enzyme spectrum data of the parameters at different time points fluctuated significantly, and the calculated V q If the value is higher than the threshold, it means that the parameter has a large deviation under different detection methods and time dimensions, which is a risk signal that requires special attention.

[0148] The square difference is (1.80-1.00) 2 =0.64, (1.95-1.05) 2 =0.81, (1.60-0.98) 2 =0.3844, (1.70-1.12) 2 = 0.3364, add the above values ​​and divide by the number of time points n = 4 and then perform square root operation, that is, Combine the two parts to get V q =0.842+0.7368=1.5788, which is the cross-validation data value of the troponin I parameter. If the risk judgment threshold is set to 1.2, the risk judgment threshold is determined based on the statistical law of the original sample data. By analyzing the cross-validation value distribution between normal samples and abnormal samples, an optimal dividing line is determined to ensure the stability and discrimination ability of the classification. Therefore, due to Vq =1.5788>1.2, so this parameter is judged as an abnormal cross-validation parameter and marked in the system to form abnormal cross-validation data;

[0149] S503: Based on the abnormal cross-validation data, a preoperative risk analysis is performed on the cross-validation parameters, and the common characteristics of the fluctuation deviation and the abnormal physiological indicators are combined to obtain a preoperative risk assessment conclusion;

[0150] According to the abnormal cross-validation data, the system needs to read the marking information of the cross-validation parameters one by one and analyze the corresponding original detection parameter data. First, the system identifies each detection indicator marked as an abnormal cross-validation parameter, and extracts the corresponding electrocardiogram ST segment standardized data and myocardial enzyme spectrum standardized data. For example, if a parameter is known to be marked as abnormal at a certain data point, and its electrocardiogram ST segment Z-score is 1.85 and its myocardial enzyme spectrum Z-score is 0.92, the system extracts the two values ​​and calculates the difference between the two, 1.85-0.92=0.93, which is the structural difference between the corresponding detection methods, and then calls the corresponding offset fluctuation abnormality value ΔA q , assuming that the value is 0.58, then call the arithmetic mean μ of the ΔA values ​​of all abnormal parameters A , set it to 0.64, complete the first half of the calculation, that is, 0.93 0.58 / 0.64 = 0.842, continue to extract the Z-score data of this parameter at multiple time points, set at 4 time points, the electrocardiogram and enzyme spectrum Z-score are [1.80, 1.95, 1.60, 1.70] and [1.00, 1.05, 0.98, 1.12] respectively, the system calculates the square of the difference between the two at each time point and averages it, that is, (0.64 + 0.81 + 0.3844 + 0.3364) / 4 = 0.5427, and then performs square root calculation Add the two parts to get the cross-validation data value V q =0.842+0.7368=1.5788, then the system sets the cross-validation index threshold to 1.2. The cross-validation index threshold is set based on the large-scale statistical analysis of the original sample data by the preoperative risk assessment model and the ROC (receiver operating characteristic) curve optimization process. The judgment operation is performed. If V q >1.2, the parameter is abnormal and is added to the preoperative risk parameter tag set. After completing the calculation of the abnormal items, the system counts the number of abnormal parameters of the data points in turn. If the number is greater than the set risk judgment value of 2, the data point is marked as a preoperative high-risk object. Otherwise, the label is not updated, and the set of data points assessed as preoperative high risk is output to form the preoperative risk assessment conclusion.

[0151] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for preoperative risk assessment in cardiology, characterized in that: The following steps are involved: S1: Obtain the patient's preoperative heart rate, blood pressure, and respiratory rate physiological parameters through medical monitoring equipment. Use a sliding window algorithm to segment the time series data, calculate the median within the window as the baseline, and offset the physiological parameter data from the baseline to obtain the fluctuation offset. S2: Based on the fluctuation offset, perform time alignment processing based on the dynamic time warping algorithm, calculate the fluctuation rate of each parameter, determine whether there is a trend abnormal change, and obtain a trend abnormality mark; S3: Calling the trend anomaly marker, extracting the corresponding fluctuation offset to evaluate the fluctuation amplitude, and comparing it with the amplitude threshold. When the fluctuation amplitudes of two or more items exceed the amplitude threshold, a risk trigger signal is generated; S4: Call the multidimensional standard deviation data in the risk trigger signal, use the K-means clustering algorithm for classification, set three cluster centers, use the Euclidean distance as the standard to calculate the cluster center value size mapping risk level, and output the risk level label; S5: Call the risk level label, detect the abnormal degree of its fluctuation offset through the isolation forest algorithm, combine the electrocardiogram ST segment and myocardial enzyme spectrum detection data to perform clinical indication cross-validation, and output the preoperative risk assessment conclusion.

2. The cardiology preoperative risk assessment method according to claim 1, characterized in that: The sliding window duration of the sliding window algorithm is 24 hours before surgery, and the window sliding step is 5 minutes; In the K-means clustering algorithm, the input data needs to be Z-score standardized before clustering, and the standard deviation screening threshold is set based on real-time data and clinical statistics to output low, medium, and high risk level labels; The trend anomaly is based on a fluctuation rate threshold set at 2 times the standard deviation of the original data mean. When the rates of two or more parameters exceed the threshold, it is determined to be a trend anomaly. The isolation forest algorithm combines the fluctuation trajectory obtained by the dynamic time warping algorithm with the abnormal period to perform spatiotemporal analysis; The fluctuation offset specifically refers to the heart rate baseline offset, blood pressure baseline offset, and respiratory rate baseline offset. The trend abnormality mark includes a unidirectional continuous trend, a bidirectional fluctuation trend, and an intermittent abnormal trend. The risk trigger signal includes a dual-parameter collaborative exceedance signal, a three-parameter joint exceedance signal, and a continuous window trigger signal. The risk level label specifically refers to a low-risk threshold interval, a medium-risk transition interval, and a high-risk warning interval. The risk assessment conclusion includes a cardiac function reserve index, a respiratory failure risk value, and a multiple organ dysfunction score.

3. The cardiology preoperative risk assessment method according to claim 1, characterized in that: The specific steps of S1 include: S101: Acquire physiological data recorded by monitoring equipment, including heart rate, blood pressure, and respiratory rate, sort the data based on timestamps, segment the time series data using a sliding window, calculate the median within the window, and generate a window benchmark median sequence; S102: Based on the window reference median sequence, call the time series data, perform offset calculation point by point, obtain the offset between the data and the window reference value, and generate a physiological parameter offset sequence; S103: extracting a large offset value, a small offset value, and a mean offset value from the offset data of the heart rate, blood pressure, and respiratory rate in the physiological parameter offset sequence, marking the corresponding monitoring times in chronological order, and calculating the offset fluctuation range of each physiological parameter to obtain a fluctuation offset; The fluctuation range of the offset is compared with the original data 72 hours before the operation for comparative analysis to obtain the baseline reference value of the movement line.

4. The cardiology preoperative risk assessment method according to claim 3, characterized in that: To calculate the offset, use the formula: Among them, Δ n Represents the offset of the nth time point, x n Represents the physiological parameter measurement value at the nth time point, M w Represents the median value of the wth window in the sliding window benchmark median sequence, μ w represents the arithmetic mean of the physiological parameter measurements within the wth sliding window, σ w represents the physiological parameter measurement value in the wth sliding window, and β is a dynamic adjustment factor determined based on the patient's age and gender.

5. The cardiology preoperative risk assessment method according to claim 3, characterized in that: The specific steps of S2 include: S201: Based on the fluctuation offset, a dynamic time warping algorithm is used to perform time series alignment processing on the two-period parameter values, identify the synchronous offset trajectory of the parameters at different time points, and generate an aligned time series feature value; S202: Calculating the change amplitude of the parameter in a continuous time period based on the aligned time series feature value, extracting the change slope of the parameter value in the time period, and comparing the change slope of each parameter with its baseline slope in the aligned time series to calculate the slope deviation of the parameter in the time period to obtain a slope offset difference value; The baseline slope is obtained by training the parameter change rate under normal conditions 72 hours before surgery; S203: Based on the slope offset difference value, a trend change identification threshold is set, and whether the parameter difference value exceeds the trend change identification threshold is determined. The parameters in the time period that continuously exceed the threshold are uniformly marked and classified to generate a trend abnormality mark; The trend change identification threshold is set with reference to clinical standards and based on expert experience.

6. The cardiology preoperative risk assessment method according to claim 5, characterized in that: The specific steps of S3 include: S301: Calling the abnormal fluctuation mark data set, extracting the fluctuation offset value of the parameter within the marked time period, obtaining the offset sequence of the parameter at the time node, fitting the offset vector to calculate the offset variance, judging the fluctuation stability based on the offset variance, and obtaining the fluctuation offset difference data; S302: Calculate the standard deviation of the parameter offset value within the marked time period based on the fluctuation offset difference data, call the amplitude threshold set by the parameter, calculate the parameter whose standard deviation value exceeds the corresponding amplitude threshold, and obtain offset amplitude threshold deviation data; S303: Calling the deviation amplitude threshold value exceeding tolerance data, setting a lower limit for determining the number of exceeding tolerances, comparing the number of marked parameters within the time period according to the lower limit, and generating a risk trigger signal based on a joint determination of the parameter deviation amplitude and the number of exceeding tolerances; The amplitude threshold is set to twice the standard deviation of the offset within the original sliding window of the parameter.

7. The cardiology preoperative risk assessment method according to claim 6, characterized in that: The specific steps of S4 include: S401: Invoke the risk trigger signal, obtain multidimensional standard deviation data for a required time period, set three cluster centers based on clustering requirements, calculate the Euclidean distance value from each data point to the three cluster centers, classify the data points into corresponding categories using the shortest distance method, and obtain the Euclidean distance mapping result; S402: Based on the Euclidean distance mapping result, extract the distance values ​​corresponding to the data points within the cluster category, use the maximum and minimum distance differences to divide the risk gradient within and between groups, set the distance range boundaries, and classify the three cluster center values ​​into three risk levels respectively, to obtain risk level classification data; S403: According to the risk level classification data, mapping and matching are performed on the corresponding relationship between each cluster label and the risk level of the cluster center, and the label classified by each data point is replaced accordingly to obtain a risk level label.

8. The cardiology preoperative risk assessment method according to claim 7, characterized in that: The Euclidean distance value is calculated using the formula: Among them, d ij Represents the Euclidean distance value from the i-th data point to the j-th cluster center, which is a dimensionless parameter, x ik Represents the standard deviation data value of the i-th data point in the k-th dimension, which is a dimensionless parameter after min-max normalization. jk Represents the coordinate value of the j-th cluster center in the k-th dimension, which is a dimensionless parameter, σ k represents the standard deviation of the k-th dimension data, which is a dimensionless parameter after min-max standardization processing, m represents the total number of data dimensions, which is a dimensionless parameter, and α represents the dynamic adjustment coefficient, which is a dimensionless parameter.

9. The cardiology preoperative risk assessment method according to claim 7, characterized in that: The specific steps of S5 include: S501: calling the risk level label, detecting the degree of abnormal deviation fluctuation based on the isolation forest algorithm, extracting the fluctuation deviation data of the abnormal parameters, filtering the abnormal fluctuations based on the detection results and marking the parameters, and obtaining the abnormal deviation fluctuation data; S502: Based on the abnormal deviation fluctuation data, according to the electrocardiogram ST segment and myocardial enzyme spectrum detection data, extract the physiological indication data of each abnormal parameter, calculate the cross-validation index according to the detection data, mark the cross-validation parameters that meet the abnormal conditions, and obtain abnormal cross-validation data; The electrocardiogram ST segment and myocardial enzyme spectrum detection data were Z-score standardized and unified; S503: Based on the abnormal cross-validation data, a preoperative risk analysis is performed on the cross-validation parameters, and a preoperative risk assessment conclusion is obtained by combining the common characteristics of fluctuation deviation and abnormal physiological indicators.

10. The cardiology preoperative risk assessment method according to claim 9, characterized in that: The cross validation index is calculated using the formula: Among them, V q Represents the cross-validation data of the qth parameter, Zs q Represents the Z-score normalized value of the electrocardiogram ST segment detection data corresponding to the qth parameter, Zm q represents the Z-score normalized value of the myocardial enzyme spectrum detection data corresponding to the qth parameter, ΔA q Represents the abnormal value of the offset fluctuation of the qth parameter, μ A Represents the arithmetic mean of the parameter deviation fluctuation anomaly, Zs q,k Represents the normalized value of the ST segment of the electrocardiogram at the kth time node, Zm q,k represents the standardized value of the myocardial enzyme spectrum of the qth parameter at the kth time node, and n represents the total number of time nodes.

Images