An aortic disease risk early warning system based on big data

Abstract

The present application relates to the technical field of risk early warning, in particular to an aortic disease risk early warning system based on big data, comprising a joint vector construction module, a distance discriminant analysis module, a misplacement offset identification module, a gradient mutation determination module and a joint risk early warning module. The present application effectively identifies the pathological state induced by electrolyte abnormalities by constructing a double-factor variation vector and combining distance discriminant analysis logic, uses multi-parameter peak position offset analysis and gradient mutation determination mechanism to deeply mine the dynamic misplacement characteristics and asynchronous change law of physiological parameters in the time dimension, constructs a misplacement matrix overlap density value set and performs matching operation with a high-risk mode, realizes accurate quantitative evaluation of the evolution trend of aortic diseases, overcomes the limitations of single static index analysis, enhances the ability to capture early hidden risk signals, and improves the sensitivity and reliability of the early warning system.

Description

Technical Field

[0001] This invention relates to the field of risk warning technology, and in particular to a big data-based aortic disease risk warning system. Background Technology

[0002] The field of risk warning technology involves analyzing, monitoring, and predicting potential risks to achieve early detection and warning of risk sources such as emergencies or diseases. It mainly includes core aspects such as data collection, risk modeling, indicator assessment, and warning triggering. This technology is widely used in various scenarios, including healthcare, natural disaster early warning, and financial risk control. By analyzing multi-source heterogeneous data, it identifies abnormal trends or high-risk states, assisting decision-makers in responding promptly. In the healthcare scenario, risk warning technology can combine multi-dimensional data such as medical indicators, physiological parameters, and past medical history to construct assessment models related to the risk of specific disease development, thereby improving individual health management.

[0003] Traditional aortic disease risk warning systems refer to systems that assess the risk of aortic-related diseases such as aortic aneurysms and aortic dissections based on static data such as historical medical records, physical examination information, and clinical examination results. These systems typically analyze whether a patient exhibits high-risk characteristics through manual judgment and rule-based comparisons. They generally rely on fixed risk scoring scales or static indicator systems built based on expert experience. Their main methods include comparing thresholds for individual physiological parameters such as blood pressure, blood lipids, and aortic diameter, combined with statistical analysis of past cases to form a judgment basis. However, they lack the ability to comprehensively analyze massive amounts of dynamic medical data and the modeling mechanism for the evolution of risk states.

[0004] Traditional early warning systems mainly rely on fixed risk scoring scales or static indicator systems built based on expert experience. They primarily use threshold comparisons of single physiological parameters such as blood pressure and aortic diameter to form judgment criteria. They lack the ability to comprehensively analyze massive amounts of dynamic medical data and cannot effectively model the evolution trend of risk states. As a result, they are unable to accurately capture subtle physiological abnormal fluctuations when faced with complex and ever-changing pathological features, causing a lag in the identification of potential high-risk states and greatly reducing the timeliness and accuracy of early warnings for sudden lesions. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a big data-based early warning system for aortic disease risk.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an aortic disease risk early warning system based on big data, the system comprising:

[0007] The joint vector construction module obtains the electrocardiogram signal and continuous serum potassium value, calculates the rate of change of high-frequency energy proportion of the electrocardiogram signal and the gradient factor of continuous serum potassium value, normalizes and splices the rate of change of high-frequency energy proportion and gradient factor, and constructs a two-factor variation vector.

[0008] The distance discriminant analysis module, based on the two-factor variation vector, calls the K-nearest neighbor algorithm to calculate its Euclidean distance with samples of similar disease states, filters vector instances whose distance is less than a preset distance radius, and generates an electrolyte abnormality warning signal;

[0009] The misalignment and offset identification module acquires diastolic blood pressure, systolic blood pressure, arterial diameter and wall stress parameters, extracts peak positions, calculates relative offset differences, and generates offset sequence curves. When the fluctuation rate of the offset sequence curve exceeds a preset fluctuation density threshold, a set of cross-abnormal points is established.

[0010] The gradient mutation determination module acquires the time series of systolic blood pressure, aortic diameter and heart rate, constructs the change rate trajectory vector and maps it to a unified time coordinate axis, calculates the time difference of local extreme points, and generates abnormal gradient combination sequences based on the time difference.

[0011] The joint risk warning module constructs a set of misalignment matrix overlap density values ​​based on the electrolyte abnormality warning signal, the cross-abnormal point set, and the abnormal gradient combination sequence, and matches them with the risk pattern to generate aortic disease risk warning results.

[0012] The present invention improves upon this invention by including the following: the dual-factor mutation vector includes normalized spectral energy coordinates, ion concentration gradient dimension, and joint feature weight coefficients; the electrolyte abnormality warning signal includes Euclidean distance metric, lesion state classification label, and abnormality severity level; the cross-abnormality point set includes peak misalignment timestamp, abnormal parameter identifier, and interval fluctuation amplitude value; the abnormal gradient combination sequence includes rate mutation value, asynchronous time difference term, and gradient chain index; and the aortic disease risk warning result includes model matching similarity score, risk evolution stage, and dominant triggering factor.

[0013] The present invention is improved in that the joint vector construction module includes:

[0014] The frequency domain feature extraction submodule acquires the electrocardiogram signal for a specified time period, performs discrete Fourier transform on the acquired time domain waveform data, maps the time domain amplitude to the frequency domain space to extract the spectral energy distribution data in the range of 0.5Hz to 40Hz, performs integral processing on the energy spectral density of the high-frequency sub-interval, calculates the proportion of the integral value to the total energy of the entire frequency band, and generates the high-frequency energy proportion change rate based on the proportion fluctuation amplitude within the continuous time window.

[0015] The ion concentration gradient calculation submodule obtains continuous blood potassium value measurement results, extracts the concentration values ​​and corresponding timestamps of adjacent sampling points in chronological order, performs a subtraction operation on the concentration values ​​of the previous and next time points to obtain the blood potassium concentration increment, calculates the time span between the corresponding sampling time points, performs a ratio operation, divides the increment value by the sampling interval duration to obtain the concentration change amplitude per unit time, and generates a gradient factor.

[0016] The vector normalization splicing submodule calls the high-frequency band energy ratio change rate and gradient factor to calculate the numerical range of each feature column. The original value is subtracted from the minimum value in the column and then divided by the range to perform normalization mapping. According to the feature dimension alignment rule, the processed energy features and gradient features are sequentially spliced ​​to construct a two-factor mutation vector.

[0017] The present invention is improved in that the distance discrimination analysis module includes:

[0018] The distance metric calculation submodule calls the two-factor mutation vector, accesses a pre-set sample library of approximate disease states, traverses the sample data in the library, calculates the numerical difference between the two-factor mutation vector and the sample in each feature dimension, performs square accumulation and arithmetic square root operations on the numerical difference, obtains the Euclidean space distance value, and generates a vector distance distribution set.

[0019] The neighborhood instance filtering submodule obtains a preset distance radius threshold based on the vector distance distribution set, compares each Euclidean space distance value in the set with the distance radius threshold, filters the nearest neighbor vector objects whose distance is less than the distance radius threshold, extracts the object identifier and establishes an index association pointing to the original data, and generates an abnormal candidate instance list.

[0020] The status determination and early warning submodule counts the cumulative number of nearest neighbor vector objects in the abnormal candidate instance list, and determines whether the number meets the minimum density requirement for lesion interval determination in combination with the time window. When it meets the requirement, it confirms that the aortic electrolyte state has entered the induced lesion interval and generates an electrolyte abnormality early warning signal.

[0021] The present invention is improved in that the process of setting the preset distance radius threshold specifically involves: obtaining a historical distribution set of aortic standard sample data in a physiologically stable state; calling a statistical analysis function to calculate the mean intra-class spatial distance and the standard deviation intra-class spatial distance between sample points in the historical distribution set; matching the corresponding statistical fold factor according to the preset risk sensitivity level; and performing an addition operation on the product of the mean intra-class spatial distance, the standard deviation intra-class spatial distance, and the statistical fold factor to obtain the preset distance radius threshold.

[0022] The process of determining whether the quantity meets the minimum density requirement for lesion interval determination by combining the time window is specifically as follows: within a preset sliding time window, frequency statistics are performed on the nearest neighbor vector objects included in the continuously generated list of abnormal candidate instances; the cumulative quantity obtained by statistics is compared with the minimum density requirement for lesion interval determination; when the cumulative quantity is greater than or equal to the minimum density requirement for lesion interval determination, it is determined that the minimum density requirement for lesion interval determination is met.

[0023] The present invention is improved in that the misalignment offset recognition module includes:

[0024] The parameter feature extraction submodule obtains parameters such as diastolic blood pressure, systolic blood pressure, arterial diameter, and wall stress. It performs extreme value search on the time-domain waveform of each parameter, locates local maxima within a continuous cardiac cycle, records the time coordinates, and generates a multidimensional peak position time series set.

[0025] The offset sequence operation submodule calls the multidimensional peak position time series set, calculates the time difference between adjacent peak points to determine the relative offset interval length difference, introduces the standard cardiac cycle duration, blood pressure pulsatility coefficient, wall stress intensity factor, arterial diameter expansion index, atrioventricular synchronization deviation and blood flow velocity gradient modulus, calculates and obtains the offset sequence intensity value, arranges the continuously calculated intensity values ​​in time order, and generates the offset sequence curve.

[0026] The abnormal pattern aggregation submodule obtains a preset physiological stable interval based on the offset sequence curve, compares the curve values ​​with the interval boundary point by point, counts the proportion of data points exceeding the boundary to calculate the fluctuation rate, compares the fluctuation rate with a preset fluctuation density threshold, and extracts the parameter combination pattern within the corresponding time period when it exceeds the fluctuation density threshold to establish a cross-abnormal point set.

[0027] The present invention is improved in that the formula for obtaining the intensity value of the offset sequence is specifically as follows:

[0028] ;

[0029] in, Represents the intensity value of the offset sequence. Represents the total number of sampling points. Representing the The difference in the relative offset interval length of each sampling point This represents the standard cardiac cycle duration. Represents the blood pressure pulsatility coefficient. Represents the wall stress intensity factor. Represents the base of the natural logarithm. Represents the arterial diameter dilation index. This represents the atrioventricular synchrony deviation. This represents the blood flow velocity gradient modulus.

[0030] The present invention is improved in that the gradient mutation determination module includes:

[0031] The rate trajectory generation submodule acquires the time series of systolic blood pressure, aortic diameter and heart rate, performs discrete difference operation on each parameter sequence, calculates the ratio of the numerical increment between adjacent sampling points to the time step, constructs a first derivative sequence characterizing the dynamic change rate of the parameters, and combines the derivative sequences of each parameter by dimension to generate a change rate trajectory vector.

[0032] The asynchronous time difference calculation submodule calls the change rate trajectory vector, uses linear interpolation logic to map each component to a unified time coordinate axis to achieve granular alignment, searches for local maxima of each component curve through a sliding window and extracts the corresponding timestamp, calculates the absolute value of the time interval between extreme points among multiple parameter types, and generates local extreme asynchronous time difference.

[0033] The gradient sequence determination submodule obtains a preset abnormal asynchronous threshold range based on the local extreme asynchronous time difference, filters time difference data that fall into the abnormal asynchronous threshold range, detects whether the time difference values ​​in multiple consecutive cardiac cycles show a monotonically increasing or decreasing evolution trend, extracts time difference data that meet the trend characteristics and their corresponding gradient indices, and generates an abnormal gradient combination sequence.

[0034] The present invention is improved in that the joint risk warning module includes:

[0035] The matrix density construction submodule calls the electrolyte abnormality warning signal, the cross-abnormal point set and the abnormal gradient combination sequence, extracts the time index and intensity parameters, maps them to a multi-dimensional feature space grid with a preset resolution, traverses all grid cells in the multi-dimensional space to count the cumulative number of abnormal feature points falling into them, calculates the ratio of the cumulative number to the grid space capacity, and generates a set of misaligned matrix overlap density values.

[0036] The pattern matching calculation submodule obtains a preset high-risk pattern density template based on the set of overlapping density values ​​of the misaligned matrix, expands the density set and template matrix into a one-dimensional numerical vector, performs multiplication and accumulation operations on the two vectors respectively to obtain the inner product value, calculates the Euclidean norm product of each vector, performs division operations to divide the inner product value by the Euclidean norm product, and generates a model matching similarity score.

[0037] The risk warning generation submodule matches the similarity score of the model, obtains a preset warning trigger threshold, compares the score with the warning trigger threshold, filters calculation instances whose scores exceed the warning trigger threshold, retrieves the corresponding lesion evolution stage description information according to the numerical range to which the score belongs, aggregates the risk level code and main triggering factors at the current moment, and generates aortic disease risk warning results.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] In this invention, by constructing a two-factor variation vector and combining it with distance discriminant analysis logic, the pathological state induced by electrolyte abnormalities can be effectively identified. By utilizing multi-parameter peak position shift analysis and gradient mutation judgment mechanism, the dynamic misalignment characteristics and asynchronous change patterns of physiological parameters in the time dimension are deeply explored. A set of misalignment matrix overlap density values ​​is constructed and matched with high-risk patterns, thereby achieving accurate quantitative assessment of the evolution trend of aortic disease. This overcomes the limitations of single static index analysis, enhances the ability to capture early hidden risk signals, and improves the sensitivity and reliability of the early warning system. Attached Figure Description

[0040] Figure 1 This is a system flowchart of the present invention;

[0041] Figure 2 This is a flowchart of the joint vector construction module of the present invention;

[0042] Figure 3 This is a flowchart of the distance discrimination analysis module of the present invention;

[0043] Figure 4 This is a flowchart of the misalignment and offset recognition module of the present invention;

[0044] Figure 5 This is a flowchart of the gradient mutation determination module of the present invention;

[0045] Figure 6 This is a flowchart of the joint risk warning module of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0047] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0048] Please see Figure 1 The present invention provides a technical solution, a big data-based aortic disease risk early warning system, including a joint vector construction module, a distance discriminant analysis module, a misalignment and offset identification module, a gradient mutation determination module, and a joint risk early warning module;

[0049] The joint vector construction module obtains the electrocardiogram signal and continuous blood potassium value measurement results for a specified time period, calls the fast Fourier transform algorithm to perform spectrum analysis on the electrocardiogram signal to calculate the rate of change of high-frequency energy proportion, calculates the concentration difference and time ratio based on the continuous blood potassium value to generate the gradient factor, performs normalization splicing operation on the rate of change of high-frequency energy proportion and the gradient factor, and constructs a two-factor variation vector.

[0050] The distance discriminant analysis module, based on the two-factor variation vector, calls the K-nearest neighbor algorithm to calculate the Euclidean distance between it and samples with similar lesion states, filters out vector instances whose distance is less than the preset distance radius, determines that the aortic electrolyte state is in the induced lesion range, and generates an electrolyte abnormality warning signal.

[0051] The misalignment and offset identification module acquires diastolic blood pressure, systolic blood pressure, arterial diameter and wall stress parameters, extracts the peak position of the parameters in a continuous cardiac cycle, calculates the difference in relative offset interval length and performs weighted merging processing to generate an offset sequence curve. Based on the comparison results between the offset sequence curve and the physiological stable interval, the fluctuation rate is calculated. When the fluctuation rate exceeds the fluctuation density threshold, the parameter offset pattern is aggregated to establish a set of cross-abnormal points.

[0052] The gradient mutation determination module acquires the time series of systolic blood pressure, aortic diameter and heart rate, calculates the first derivative rate of change to generate a rate of change trajectory vector, maps the rate of change trajectory vector to a unified time coordinate axis, calculates the time difference of local extreme points between parameters, and determines that when the time difference falls into the abnormal asynchronous threshold range and has sequentiality, it generates an abnormal gradient combination sequence.

[0053] The joint risk warning module constructs a set of misaligned matrix overlap density values ​​based on electrolyte abnormality warning signals, cross-abnormal point sets, and abnormal gradient combination sequences. It then uses a cosine similarity algorithm to calculate the matching degree between the misaligned matrix overlap density value set and the high-risk pattern density template. When the matching degree is higher than the warning trigger line, it generates an aortic disease risk warning result.

[0054] The two-factor variation vector includes normalized spectral energy coordinates, ion concentration gradient dimension, and joint feature weight coefficients. The electrolyte abnormality warning signal includes Euclidean distance metric, lesion status classification label, and abnormality severity level. The cross-abnormality point set includes peak misalignment timestamp, abnormal parameter identifier, and interval fluctuation amplitude value. The abnormal gradient combination sequence includes rate mutation value, asynchronous time difference term, and gradient chain index. The aortic disease risk warning result includes model matching similarity score, risk evolution stage, and dominant triggering factor.

[0055] Please see Figure 2 The joint vector construction module includes:

[0056] The frequency domain feature extraction submodule acquires the electrocardiogram signal for a specified time period, performs discrete Fourier transform on the acquired time domain waveform data, maps the time domain amplitude to the frequency domain space to extract the spectral energy distribution data in the range of 0.5Hz to 40Hz, performs integral processing on the energy spectral density of the high-frequency sub-interval, calculates the proportion of the integral value to the total energy of the entire frequency band, and generates the high-frequency energy proportion change rate based on the proportion fluctuation amplitude within the continuous time window.

[0057] The process of generating the high-frequency energy proportion change rate based on the proportion fluctuation amplitude within a continuous time window is as follows: the time length of the sliding window is set, which is dynamically matched according to the average cardiac cycle duration of the subjects; the variance of the proportion values ​​at multiple consecutive moments within the sliding window is calculated, and this variance is used as the high-frequency energy proportion change rate.

[0058] The device is directly connected to a bioelectrical signal acquisition probe via a high-speed data bus. This probe incorporates a high-sensitivity analog front-end circuit, configured to continuously capture ECG voltage changes on the subject's body surface at a sampling frequency of 1000 Hz. It integrates a signal preprocessing unit, which first performs noise reduction on the received raw time-domain waveform data using a Butterworth bandpass filter of order 4, with passband cutoff frequencies set at 0.05 Hz and 100 Hz to filter out baseline drift noise and power frequency interference. Subsequently, the built-in digital signal processing core performs a discrete Fourier transform on the time-domain waveform data within a specified time period (set as a continuous 30-second monitoring window). This computational process employs a radix-2 Fast Fourier Transform (FFT) algorithm to convert a discrete time-domain sequence of length N into a complex frequency-domain sequence of the same length. By calculating the modulus of the complex sequence, the time-domain amplitude is precisely mapped to the frequency domain, thereby generating spectral energy distribution data. Built-in spectral filtering logic automatically locks and extracts spectral components within a specific frequency band of 0.5 Hz to 40 Hz, covering the main energy concentration areas of the P wave, QRS complex, and T wave in ECG signals. Further high-frequency energy integration is then performed. Based on a preset frequency band division threshold (set to 15 Hz), the extracted spectral data is divided into low-frequency components (0.5 Hz to 15 Hz) and high-frequency components (15 Hz to 40 Hz). For the high-frequency components, each discrete frequency sampling point within this interval is traversed, and the spectral amplitude (in microvolts) at the current frequency point is read. The square of the calculated spectral amplitude is then integrated, and the results from all sampling points within the high-frequency component are continuously accumulated using an accumulation register. Finally, the energy spectral density integral value of the high-frequency sub-interval is output. To quantify the weight of high-frequency energy in the overall spectrum, the total energy integral across the entire frequency band from 0.5 Hz to 40 Hz was calculated simultaneously. The energy spectral density integral value of the high-frequency sub-interval was divided by the total energy integral value across the entire frequency band to obtain a dimensionless weight value. To capture subtle fluctuations in the early stages of disease, a dynamic sliding window mechanism was configured, with the window duration set to 10 times the average cardiac cycle length of the subject over the past 24 hours (e.g., if the average cardiac cycle is 0.8 seconds, the window length is 8 seconds). Within each sliding window, 10 consecutively generated weight values ​​were cached, and the variance calculation unit was invoked to calculate the statistical variance of these 10 values. This variance result directly characterizes the dispersion of the high-frequency energy proportion, outputting it as the rate of change of the high-frequency energy proportion. This parameter can sensitively reflect the high-frequency oscillation characteristics of myocardial cell electrophysiological activity under electrolyte disturbances, providing core frequency domain indicators for subsequent abnormal pattern recognition.

[0059] The ion concentration gradient calculation submodule obtains continuous blood potassium value measurement results, extracts the concentration values ​​and corresponding timestamps of adjacent sampling points in chronological order, performs a subtraction operation on the concentration values ​​of the previous and next time points to obtain the blood potassium concentration increment, calculates the time span between the corresponding sampling time points, performs a ratio operation, divides the increment value by the sampling interval duration to obtain the concentration change amplitude per unit time, and generates a gradient factor characterizing the dynamic evolution rate of ion concentration.

[0060] The process of performing ratio calculation, which divides the incremental value by the sampling interval, specifically involves determining whether the time span between the corresponding sampling moments meets the minimum interval constraint. The minimum interval constraint is determined based on the electrochemical reaction stabilization time of the blood potassium testing instrument. If it meets the constraint, the division operation is performed to obtain the instantaneous gradient value; otherwise, the current data is treated as noise and discarded.

[0061] The data output of the analyzer connects to an implantable or wearable microfluidic biochemical analyzer via a physical interface. Equipped with a potassium-selective electrode, the analyzer outputs the current serum potassium concentration (mmol / L) measurement every 5 minutes. Internally, it features a circular data buffer to store the latest received serum potassium concentration values ​​and their corresponding millisecond-level timestamps in real-time, chronological order. When a new measurement result is stored in the buffer, differential calculation logic is immediately triggered to extract the current concentration value. Concentration value compared to the previous sampling time Perform subtraction ( This allows us to obtain the increase in hemostatic potassium concentration. Simultaneously, the timestamps corresponding to the two data points are read. and The absolute difference between the two values ​​is calculated as the time span. To ensure the validity of the gradient calculation and eliminate spurious fluctuations caused by instrument instability, a strict minimum interval constraint logic is built in. This constraint is based on the minimum physical time (set to 120 seconds) required for the electrochemical reaction in the serum potassium analyzer to reach steady state. The calculated time span is compared with 120 seconds. Only when the time span is greater than or equal to 120 seconds is the current sampling interval considered valid, and then a ratio calculation is performed, dividing the increase in serum potassium concentration by the time span (converted to minutes) to obtain the concentration change amplitude per unit time. For example, if the concentration increase is 0.3 mmol / L and the time span is 5 minutes, the calculated gradient value is 0.06 mmol / L / min. If the time span is less than 120 seconds, the current data is judged as transient noise or measurement artifacts, and an automatic rejection operation is performed, without gradient update, and the gradient value from the previous moment is retained. This process generates a gradient factor that characterizes the rate of dynamic change in ion concentration. This factor not only reflects the absolute change in blood potassium levels but also reveals the severity of electrolyte disturbances. It has key early warning value for identifying aortic wall stress abnormalities induced by acute hyperkalemia or hypokalemia.

[0062] The vector normalization splicing submodule calls the change rate of high-frequency energy proportion and gradient factor to calculate the numerical range of each feature column. The original value is subtracted from the minimum value in the column and then divided by the range to perform normalization mapping. According to the feature dimension alignment rule, the processed energy features and gradient features are sequentially spliced ​​to construct a two-factor mutation vector.

[0063] The process of calculating the numerical range of each feature column is as follows: in the dynamic data stream, the historical maximum and minimum values ​​of the change rate of the energy proportion of the high-frequency band and the gradient factor are updated in real time, and the difference between the historical maximum and the historical minimum values ​​is used as the normalized denominator.

[0064] Internally, two independent dynamic extreme value register sets are maintained, corresponding to energy features and gradient features respectively. In the continuous dynamic data stream, the current input value is compared in real time with the historical maximum and minimum values ​​stored in the registers. Once the current value exceeds the historical maximum or falls below the historical minimum, the registers are immediately updated to ensure that the extreme value range always covers the full range of fluctuations in the current monitoring period. Based on the real-time updated extreme values, the numerical range of each feature dimension is calculated, i.e., the current historical maximum minus the historical minimum is used as the normalized denominator. Subsequently, a linear normalization mapping operation is performed: for the rate of change of the high-frequency energy proportion of the current input... ,calculate For the current gradient factor ,calculate This process maps two features with completely different physical dimensions (the former being dimensionless variance, and the latter being concentration rate) uniformly into a closed interval of 0 to 1, eliminating weight bias caused by the difference in magnitude. After normalization, a serial splicing operation is performed according to a preset feature dimension alignment rule. This rule defines a one-dimensional vector container of length 2, placing the processed energy feature value at index 0 and the gradient feature value at index 1, thereby constructing a standardized two-factor mutation vector. This vector accurately represents the coupling state of cardiac electrophysiological abnormalities and blood biochemical abnormalities within the same mathematical space, providing a standardized input format for subsequent multidimensional spatial distance calculations.

[0065] Please see Figure 3 The distance discrimination analysis module includes:

[0066] The distance metric calculation submodule calls the two-factor mutation vector, accesses the pre-set sample library of approximate disease states, traverses the sample data in the library, calculates the numerical difference between the two-factor mutation vector and the sample in each feature dimension, performs square accumulation and arithmetic square root operation on the numerical difference, obtains the Euclidean space distance value representing the geometric similarity of multidimensional space, and generates a set of vector distance distributions.

[0067] The system directly accesses a sample library of approximate disease states stored in non-volatile memory. This library contains 1000 sets of historical two-factor vector data representing the early stages of clinically diagnosed aortic lesions. Each set of data represents a typical combination of pathological features. Upon receiving the real-time constructed two-factor variance vector, the system immediately initiates a traversal calculation engine to process each sample vector in the library. Calculate each one with the real-time vector The numerical deviation across two feature dimensions. Specifically, the difference in the energy dimension is calculated. Difference with gradient dimension Then, the floating-point arithmetic unit is invoked to perform a squaring operation on the two differences ( and The results are accumulated, and the arithmetic square root of the accumulated result is taken to obtain the standard Euclidean distance value. This calculation process strictly follows the principle of multidimensional spatial geometric similarity. The smaller the calculated distance value, the higher the similarity between the current patient's physiological state and the lesion sample in the database. These 1000 distance calculation tasks are processed in parallel, and the 1000 generated Euclidean distance values ​​are organized into a vector distance distribution set, with the corresponding index ID in the sample database attached, and output to the next-level filtering module. Through large-scale parallel computing capabilities, the abstract physiological parameters are mapped in real time to a visualized lesion spatial distance distribution, laying a quantitative foundation for similarity-based diagnostic reasoning.

[0068] The neighborhood instance filtering submodule obtains a preset distance radius threshold based on the vector distance distribution set, compares each Euclidean space distance value in the set with the distance radius threshold, filters the nearest neighbor vector objects whose distance is less than the distance radius threshold, extracts the object identifier and establishes an index association pointing to the original data, and generates an abnormal candidate instance list.

[0069] The process of setting the preset distance radius threshold is as follows: obtain the historical distribution set of aortic standard sample data in a physiologically stable state; call the statistical analysis function to calculate the mean and standard deviation of intra-class spatial distance between sample points in the historical distribution set; match the corresponding statistical fold factor according to the preset risk sensitivity level; perform addition operation on the product of the mean intra-class spatial distance, the standard deviation of intra-class spatial distance, and the statistical fold factor to obtain the preset distance radius threshold;

[0070] To accurately identify high-risk individuals from massive amounts of distance data, the system first retrieves a preset distance radius threshold by accessing the system configuration register. This threshold is based on rigorous statistical analysis: the system pre-imports a historical distribution set of 5000 standard aortic samples in a physiologically stable state (no aortic disease and normal electrolytes), calls statistical analysis functions to calculate the pairwise intra-class spatial distance between all sample points in this set, yielding a mean intra-class spatial distance of 0.12 and a standard deviation of 0.03. Based on the current risk sensitivity level (set to "high sensitivity" mode), the system matches the corresponding statistical fold factor of 3 and performs an addition operation: A preset distance radius threshold of 0.21 is set. This threshold represents the limit of vector fluctuation under normal physiological conditions within a 99.7% confidence interval. During operation, the vector distance distribution set input by the distance metric calculation submodule is traversed, and each Euclidean distance value in the set is compared with 0.21. Samples with a distance value less than 0.21 are determined to fall into the "lesion nearest neighbor region" of the current real-time vector, that is, the current state is highly similar to the lesion sample. All nearest neighbor vector objects that meet the conditions are screened out, their unique object identifiers in the sample library are extracted, and an index association pointing to the original clinical data (such as diagnosis time, lesion type) is established. Finally, an abnormal candidate instance list containing all high-risk matching items is generated.

[0071] The status determination and early warning submodule counts the cumulative number of nearest neighbor vector objects in the list of abnormal candidate instances, and determines whether the number meets the minimum density requirement for lesion interval determination in combination with the time window. When it meets the requirement, it confirms that the aortic electrolyte status has entered the induced lesion interval and generates an electrolyte abnormality early warning signal.

[0072] The specific method for determining the minimum density requirement for lesion interval determination is as follows: obtain the total number of neighborhood samples selected in the K-nearest neighbor algorithm; obtain the risk confidence ratio coefficient obtained by back-inference based on historical false alarm data; perform a multiplication operation on the total number of neighborhood samples selected and the risk confidence ratio coefficient, and perform an up-rounding operation on the result to obtain the lower limit value of the quantity that characterizes the effectiveness of abnormal clustering, and establish the lower limit value as the minimum density requirement for lesion interval determination;

[0073] The process of determining whether the number of cases meets the minimum density requirement for lesion interval determination by combining the time window is as follows: within a preset sliding time window, frequency statistics are performed on the nearest neighbor vector objects contained in the continuously generated list of abnormal candidate instances; the cumulative number obtained by statistics is compared with the minimum density requirement for lesion interval determination; when the cumulative number is greater than or equal to the minimum density requirement for lesion interval determination, it is determined that the minimum density requirement for lesion interval determination is met.

[0074] The K-Nearest Neighbors (KNN) algorithm is used to perform density analysis on the list of abnormal candidate instances. First, a minimum density requirement for lesion region determination is obtained. This requirement is determined as follows: The total number of neighboring samples selected in the KNN algorithm, K (set to 50), is obtained; the risk confidence ratio coefficient (set to 0.6, meaning at least 60% of the neighbors must be lesion samples) is obtained based on back-calculation of historical false alarm data; and a multiplication operation is performed. The result is rounded up (if the result is a decimal), and 30 is established as the lower limit for the number. In real-time monitoring, a sliding time window with a time span of 5 minutes is initiated. Within this window, the number of nearest neighbor vector objects accumulated in the list of abnormal candidate instances is continuously counted. An internal counter updates the count value in real time and compares it with the minimum density requirement of 30. When the accumulated number reaches or exceeds 30, it means that known lesion samples are densely distributed around the current physiological state feature space, and the aggregation density exceeds the statistical significance lower limit. At this time, it is determined that the aortic electrolyte state has entered the induced lesion zone, the risk warning flag is immediately activated (set to high level), and an electrolyte abnormality warning signal containing a timestamp and density value is generated. This signal, as a primary trigger source, directly leads to the final risk decision unit, ensuring an early warning before electrolyte disturbances cause structural lesions.

[0075] Please see Figure 4 The misalignment detection module includes:

[0076] The parameter feature extraction submodule obtains parameters such as diastolic blood pressure, systolic blood pressure, arterial diameter, and wall stress. It performs extreme value search on the time-domain waveform of each parameter, locates local maxima within a continuous cardiac cycle, records the time coordinates, and generates a multidimensional peak position time series set.

[0077] The system connects to multi-channel physiological parameter monitoring equipment to acquire real-time data streams of diastolic blood pressure (from invasive arterial pressure monitoring or continuous cuff measurement), systolic blood pressure, arterial diameter (from M-mode ultrasound real-time tracking), and wall stress parameters (estimated through wall thickness and pressure values). An internal extreme value search logic unit performs sliding window extreme value detection within a continuous cardiac cycle waveform for each parameter channel. For example, for the arterial diameter waveform, the logic unit compares the current sampling point with the values ​​of the five adjacent sampling points. If the current value is greater than all surrounding points, it is identified as a local maximum, and the precise value of this maximum and its corresponding millisecond-level time coordinate are recorded. By processing multiple consecutive cardiac cycles, sequences of peak systolic blood pressure, peak diastolic blood pressure, maximum arterial diameter, and maximum wall stress are generated. These discrete time points and amplitude data are encapsulated into a multidimensional peak position time series set, providing precise time-domain anchors for subsequent analysis of the phase lag relationship between various parameters.

[0078] The offset sequence calculation submodule calls a multidimensional peak position time series set to calculate the time difference between adjacent peak points to determine the relative offset interval length difference. It incorporates standard cardiac cycle duration, blood pressure pulsatility coefficient, wall stress intensity factor, arterial diameter dilation index, atrioventricular synchronization deviation, and blood flow velocity gradient modulus, using the following formula:

[0079] ;

[0080] The offset sequence intensity values ​​are obtained through calculation, and the continuously calculated intensity values ​​are arranged in chronological order to generate the offset sequence curve.

[0081] in, Represents the intensity value of the offset sequence. This represents the total number of sampling points, obtained by counting the number of discrete data points included within the current analysis time window. Representing the The difference in the relative offset interval length of each sampling point is obtained by calculating the absolute difference between the timestamp of the current peak feature point and the timestamps of the adjacent peak feature points. The standard cardiac cycle duration is represented by the RR interval extracted from the user's historical ECG data and its arithmetic mean calculated. The blood pressure pulsatility coefficient is obtained by calculating the variance of the systolic blood pressure sequence and then normalizing it by dividing it by a preset standard physiological baseline variance. The wall stress intensity factor is obtained by collecting instantaneous wall stress values ​​and dividing them by a preset aortic wall ultimate tensile strength constant to obtain a dimensionless ratio. Represents the base of the natural logarithm, and is a mathematical constant. The arterial diameter dilation index is obtained by measuring the relative rate of change of arterial diameter during cardiac ejection relative to its end-diastolic diameter. The atrioventricular synchrony deviation is represented by the phase difference between the P wave and the R wave, which is then normalized and compared according to the standard PR interval. The blood flow velocity gradient modulus is obtained by calculating the spatial derivative of the blood flow velocity field and normalizing it using the maximum theoretical flow velocity value.

[0082] The multidimensional peak position time series is called to calculate the time difference between adjacent peak points (such as the time of peak systolic blood pressure and the time of maximum arterial diameter) and determine the difference in relative offset interval length. Simultaneously, standard cardiac cycle durations are introduced from databases or real-time streams. Blood pressure pulsatility coefficient Wall stress intensity factor Arterial diameter dilation index atrioventricular synchrony deviation and blood flow velocity gradient modulus It uses a dedicated arithmetic logic unit to perform formula calculations.

[0083] The specific acquisition and quantification process of each parameter involved in the formula is as follows:

[0084] Blood pressure pulsatility coefficient The submodule collects the systolic blood pressure sequence over the past 5 minutes and calculates its variance. Let the variance of the current systolic blood pressure sequence be 150 mmHg. The preset standard physiological baseline variance (resting state in healthy individuals) is 100 mm. Perform normalized division: ,Right now This coefficient reflects the degree of fluctuation in blood pressure;

[0085] Wall stress intensity factor Instantaneous wall stress values ​​are collected using sensors, and the peak wall stress is set to 200 kPa. The preset ultimate tensile strength constant of the aortic wall is 500 kPa. Dimensionless ratios are calculated. ,Right now This factor characterizes the extent to which the physical load borne by the blood vessel wall approaches its limit;

[0086] Arterial diameter dilation index The module measures the maximum arterial diameter during cardiac ejection (e.g., 22 mm) and the minimum diameter at end-diastole (e.g., 20 mm). It calculates the relative rate of change. ,Right now ;

[0087] atrioventricular synchrony deviation By synchronizing ECG signals, the time difference between the P-wave onset and the R-wave peak (PR interval) was extracted, and the measured value was set to 0.18 seconds. The standard PR interval duration was set to 0.16 seconds. Normalized comparison was then performed. ,Right now ;

[0088] Blood flow velocity gradient modulus The spatial derivative of the blood flow velocity field is calculated using Doppler ultrasound data, assuming the current gradient magnitude is 50. The gradient baseline value corresponding to the maximum theoretical flow velocity is 200. Normalized mapping: ,Right now ;

[0089] Based on the above parameters, execute the formula:

[0090] ;

[0091] Table 1. Calculation Examples of Offset Sequence Strength Values

[0092]

[0093] Empirical evidence of the calculation process:

[0094] First, calculate the first part inside the square root: ;

[0095] Calculate the second part inside the square root: ;

[0096] Sum of all items inside the square root: ;

[0097] Square root: ;

[0098] Calculate the subtrahend: ;

[0099] Single-point calculation results: ;

[0100] All within the window Each point performs the above calculation and accumulates the results (only a single-point logic is shown here), ultimately generating... The value comprehensively reflects the degree of decoupling between the mechanical response of the vessel wall and the cardiac electrical signal under a specific hemodynamic load. The technical advantage of this formula lies in the introduction of an exponential term. This nonlinearly amplifies the sensitivity of arterial dilation to wall stress, allowing for a significant jump in calculated intensity values ​​even in the early stages of minute pathological dilation, thus achieving highly sensitive monitoring and enabling continuously calculated values ​​to be used in conjunction with wall stress. The values ​​are arranged in chronological order to generate an offset sequence curve.

[0101] The abnormal pattern aggregation submodule obtains the preset physiological stable interval based on the offset sequence curve, compares the curve value with the interval boundary point by point, counts the proportion of data points exceeding the boundary to calculate the fluctuation rate, compares the fluctuation rate with the preset fluctuation density threshold, and extracts the parameter combination pattern within the corresponding time period when it exceeds the fluctuation density threshold to establish a cross-abnormal point set.

[0102] The specific method for setting the fluctuation density threshold is as follows: acquire the offset sequence data of patients with historical aortic disease and healthy control groups, construct the receiver operating characteristic curve that distinguishes the two groups of samples, select the fluctuation rate value corresponding to the maximum point of Youden's index as the initial threshold, fine-tune the initial threshold in combination with the upper limit of the allowed false positive rate, and determine the fine-tuned value as the fluctuation density threshold.

[0103] The process of obtaining the preset physiological stability interval is as follows: extract the statistical distribution characteristics of the offset sequence of healthy people in a large-scale sample, calculate the mean and standard deviation of the distribution, and expand the range to both sides by three times the standard deviation with the mean as the center, and define the range as the physiological stability interval.

[0104] The process of calculating the fluctuation rate by counting the proportion of data points that exceed the boundary is as follows: traverse each discrete data point on the offset sequence curve, determine whether the ordinate value of each point falls outside the physiological stability range, and if it does, accumulate the abnormal count. Divide the accumulated abnormal count by the total number of points in the current statistical window to generate the fluctuation rate.

[0105] Based on the principles of statistical process control, the offset sequence curve was analyzed. First, a predefined physiological stability interval was defined. This interval was obtained based on large-scale healthy population data, and the calculated mean of the offset sequence distribution was 50, with a standard deviation of 5. (Based on 3...) The principle is to expand outwards from the mean value to determine a stable interval of [35, 65]. Then, iterate through each discrete data point on the offset sequence curve, checking if its ordinate value falls outside the [35, 65] interval. For example, if at a certain moment... A value of 70 is considered an outlier, and the counter is incremented by 1. If 150 outliers are found within a statistical window containing 1000 data points, the module calculates the fluctuation rate. (i.e., 15%). This fluctuation rate is then compared to a preset fluctuation density threshold. This threshold is set using the Receiver Operating Characteristic (ROC) curve method: analyzing historical fluctuation rate data from patients and healthy individuals, the value corresponding to the maximum point of the Youden index (sensitivity + specificity - 1), 0.12, is selected as the initial threshold. Considering clinical tolerance for false positives and the allowable upper limit of the false positive rate (5%), the initial threshold is fine-tuned to 0.13. In the current example, 0.15 > 0.13, exceeding the threshold. Therefore, an abnormal pattern cluster is identified within this time period. All parameter combinations within this period are extracted to establish a set of cross-outliers, which are then marked as pathological data clusters requiring focused attention.

[0106] Please see Figure 5 The gradient mutation determination module includes:

[0107] The rate trajectory generation submodule acquires the time series of systolic blood pressure, aortic diameter and heart rate, performs discrete difference operation on each parameter sequence, calculates the ratio of the numerical increment between adjacent sampling points to the time step, constructs a first derivative sequence characterizing the dynamic change rate of the parameters, and combines the derivative sequences of each parameter by dimension to generate a change rate trajectory vector.

[0108] The system captures the dynamic trends of physiological parameters, acquiring three time-series data streams in parallel: systolic blood pressure, aortic diameter, and heart rate. For each sequence, a first-order discrete-difference operation is performed. Taking the systolic blood pressure sequence as an example, if... The reading was 120 mmHg. The time value is 122 mmHg, the sampling time step is 0.5 seconds, then the calculated value increment is 2, and the ratio is... This process is performed point-by-point on all parameter sequences in mmHg / s to construct sequences of systolic blood pressure change rate, diameter change rate, and heart rate change rate, respectively. Then, following the format [systolic blood pressure change rate, diameter change rate, heart rate change rate], the three derivative values ​​at the same moment are combined to generate a multidimensional rate of change trajectory vector. This vector depicts the speed and direction of the evolution of the cardiovascular system state in phase space, revealing specific dynamic characteristics such as "a sharp rise in blood pressure accompanied by a slow expansion of the diameter."

[0109] The asynchronous time difference calculation submodule calls the rate of change trajectory vector and uses linear interpolation logic to map each component to a unified time coordinate axis to achieve granular alignment. It searches for local maxima of each component curve through a sliding window and extracts the corresponding timestamps. It calculates the absolute value of the time interval between extreme points among multiple parameter types and generates local extreme asynchronous time differences.

[0110] Since different physiological signals may be acquired at different frequencies (e.g., ECG at 1000Hz, blood pressure at 100Hz), a linear interpolation algorithm is first used to uniformly map all components of the rate of change trajectory vector onto a 1000Hz time axis, achieving granular alignment. After alignment, a sliding window (2-second window width) is activated to search for local maxima in each component curve. For example, within a certain cardiac cycle, the extreme point of the systolic blood pressure rate of change is detected at... Seconds, and the extreme point of the rate of change of aortic diameter appears at... seconds, calculate the absolute value of the time interval between the two. This calculation is performed on all parameter pairs every 60 milliseconds (seconds) to generate a set of asynchronous time differences for local extrema. This metric reflects the hysteresis effect of the vascular wall's mechanical response relative to hemodynamic drives and is an important physical quantity for assessing changes in vascular wall compliance.

[0111] The gradient sequence determination submodule obtains a preset abnormal asynchronous threshold range based on the local extreme asynchronous time difference, filters the time difference data that fall into the abnormal asynchronous threshold range, detects whether the time difference value in multiple consecutive cardiac cycles shows a monotonically increasing or decreasing evolution trend, extracts the time difference data that meets the trend characteristics and its corresponding gradient index, and generates an abnormal gradient combination sequence.

[0112] The specific process for obtaining the abnormal asynchronous threshold interval is as follows: First, a pre-set aortic lesion sample set and a physiologically healthy control sample set are obtained. Then, the time interval distribution data of extreme points between the differential parameter types in the two sample sets are extracted. Statistical analysis is performed on the time interval distribution data of the physiologically healthy control sample set to calculate the arithmetic mean and standard deviation of the data, constructing a reference model that conforms to a normal distribution. The upper bound of the interval of the reference model at a pre-set confidence level is selected and defined as the physiological safety baseline. Density-based clustering is performed on the time interval distribution data of the aortic lesion sample set to identify core lesion feature clusters and determine the lower limit of the values ​​for each cluster. The difference between the physiological safety baseline and the lower limit is calculated, multiplied by a pre-set safety margin coefficient, and added to the physiological safety baseline to obtain the starting threshold of the abnormal asynchronous threshold interval. Finally, the maximum statistical value of the time interval distribution data in the aortic lesion sample set is set as the ending threshold of the abnormal asynchronous threshold interval.

[0113] The abnormal asynchronous threshold interval was determined as follows: 1000 physiologically healthy control samples were collected, and the extreme time intervals between parameters were extracted, yielding a mean of 30 ms and a standard deviation of 5 ms. A normal distribution model was constructed, and the upper bound of the interval (approximately 40 ms) at a 95% confidence level was selected as the physiological safety baseline. 500 aortic lesion samples were collected, and through DBSCAN clustering, the lower limit of the time interval for identifying core lesion clusters was determined to be 55 ms. The difference was calculated. ms, multiplied by a safety margin factor of 0.8, equals 12ms. The initial threshold is set to... The termination threshold is set to the maximum value of the lesion sample, 200ms. That is, the abnormal asynchronous threshold range is [52ms, 200ms]. During monitoring, time difference data falling within the range of [52ms, 200ms] are selected. If a monotonically increasing trend in time difference values ​​is detected over five consecutive cardiac cycles (e.g., 55ms, 58ms, 62ms, 67ms, 73ms), this is considered a characteristic signal of increased vascular wall hardening. These time difference data and their corresponding gradient indices are extracted to generate an abnormal gradient combination sequence, which serves as input for subsequent risk assessment.

[0114] Please see Figure 6 The joint risk warning module includes:

[0115] The matrix density construction submodule calls the electrolyte abnormality warning signal, the cross-abnormal point set and the abnormal gradient combination sequence, extracts the time index and intensity parameters, maps them to the multi-dimensional feature space grid of the preset resolution, traverses all grid cells in the multi-dimensional space to count the cumulative number of abnormal feature points falling into them, calculates the ratio of the cumulative number to the grid space capacity, and generates a set of misaligned matrix overlap density values.

[0116] The process of mapping it to a multi-dimensional feature space grid with a preset resolution is as follows: a three-dimensional orthogonal coordinate system containing time evolution dimension, signal intensity amplitude dimension, and spatial position dimension is established; the three-dimensional orthogonal coordinate system is discretized according to the preset time sampling interval, signal quantization ladder, and anatomical partition grid to construct a three-dimensional feature space composed of multiple discrete grid units; the time index is mapped to the corresponding discrete interval of the time evolution dimension, the intensity parameter is mapped to the corresponding quantization level of the signal intensity amplitude dimension, and combined with the preset spatial positioning mark, it is mapped to the spatial position dimension, thereby determining the specific coordinate position of each data point in the multi-dimensional feature space grid;

[0117] To calculate risk density, multi-source anomaly features are fused into a unified three-dimensional space. First, a three-dimensional orthogonal coordinate system is established: the X-axis represents the time evolution dimension (discrete interval of 1 minute), the Y-axis represents the signal intensity amplitude dimension (quantized into 10 levels), and the Z-axis represents the spatial location dimension (corresponding to anatomical regions, such as the ascending aorta and aortic arch). Electrolyte anomaly warning signals, cross-anomaly point sets, and anomaly gradient combination sequences are retrieved, and the time index (e.g., 10:05), intensity parameter (e.g., 0.8), and spatial location marker (e.g., Zone2) are extracted. According to mapping rules, data points are projected onto corresponding grid cells. For example, if a data point falls into the grid at coordinates (10:05, Level8, Zone2), the entire multi-dimensional feature space grid is traversed, and the cumulative number of anomaly feature points falling into each grid cell is counted. If a grid space is designed for 100 points, and 40 points have been accumulated, the ratio is calculated as follows: This value is the overlap density value of the misalignment matrix, which ultimately generates a set containing all non-zero grid density values. This set visually demonstrates the degree of clustering of lesion features at a specific time, intensity, and location.

[0118] The pattern matching calculation submodule obtains a preset high-risk pattern density template based on the set of overlapping density values ​​of the misaligned matrix. It expands the density set and the template matrix into a one-dimensional numerical vector, performs multiplication and accumulation operations on the two vectors to obtain the inner product value, calculates the Euclidean norm product of each vector, performs division operations to divide the inner product value by the Euclidean norm product, and generates a model matching similarity score.

[0119] Obtain a pre-set high-risk pattern density template, which is a standard matrix trained on a large number of confirmed cases. Expand both the input set of overlapping density values ​​of the misaligned matrix and the template matrix into one-dimensional numerical vectors. Let the density set vector be... The template vector is Perform dot product operation And calculate the Euclidean norm respectively. and Finally, perform the division operation. This generates a model matching similarity score in the form of cosine similarity (ranging from 0 to 1). A score of 0.95 indicates that the current patient's feature distribution closely matches the high-risk pattern.

[0120] The risk warning generation submodule obtains the preset warning trigger threshold based on the model matching similarity score, compares the score with the warning trigger threshold, filters the calculation instances whose scores exceed the warning trigger threshold, retrieves the corresponding lesion evolution stage description information according to the numerical range to which the score belongs, aggregates the risk level code and main triggering factors at the current moment, and generates aortic disease risk warning results.

[0121] The process of obtaining the preset warning trigger threshold is as follows: First, obtain a historical validation dataset containing confirmed positive samples and excluded negative samples. Second, calculate the corresponding model matching similarity score for each sample in the historical validation dataset and construct a histogram of the probability distribution of scores for positive and negative samples. Third, traverse the score intervals in the probability distribution histogram and calculate the sensitivity and specificity indices at different boundary points. Fourth, select the boundary point value where the difference between the sum of the sensitivity and specificity indices and one reaches its maximum as the basic threshold. Fifth, obtain a preset maximum allowable false alarm rate parameter. When the actual false alarm rate calculated based on the basic threshold exceeds the maximum allowable false alarm rate parameter, adjust the basic threshold along the direction of increasing score until the false alarm rate constraint is met, and set the adjusted value as the preset warning trigger threshold.

[0122] Risk levels are determined based on model matching similarity scores, requiring the acquisition of a preset warning trigger threshold. This threshold is obtained from a historical validation dataset (containing 500 confirmed positive cases and 500 excluded negative cases). A histogram is constructed by calculating the matching score for each sample and iterating through the score range (0.0 to 1.0, with a step size of 0.01). Sensitivity and specificity are calculated at each cutoff point. For example, at the cutoff point of 0.75, the sensitivity is 0.90 and the specificity is 0.85; the sum of these two values ​​minus one equals 0.75. Subsequently, the false alarm rate calculated based on 0.78 is checked. If the actual false alarm rate is 8%, and the preset maximum allowable false alarm rate parameter is 5%, the base threshold is adjusted in the direction of increasing score (e.g., adjusted to 0.82) until the false alarm rate drops below 5%. Finally, 0.82 is set as the preset warning trigger threshold. In real-time monitoring, if the current model matching similarity score (e.g., 0.85) exceeds 0.82, the corresponding disease evolution stage description information (e.g., "aortic dissection prodromal stage") is retrieved, the current risk level code (Level 4) and the main triggering factors (e.g., "high-frequency electrolyte oscillation + abnormal vascular wall compliance gradient") are aggregated, the final aortic disease risk warning result is generated, and pushed to the doctor's workstation or the patient's mobile terminal.

[0123] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A big data-based aortic disease risk early warning system, characterized in that, The system includes: The joint vector construction module obtains the electrocardiogram signal and continuous serum potassium value, calculates the rate of change of high-frequency energy proportion of the electrocardiogram signal and the gradient factor of continuous serum potassium value, normalizes and splices the rate of change of high-frequency energy proportion and gradient factor, and constructs a two-factor variation vector. The distance discriminant analysis module, based on the two-factor variation vector, calls the K-nearest neighbor algorithm to calculate its Euclidean distance with samples of similar disease states, filters vector instances whose distance is less than a preset distance radius, and generates an electrolyte abnormality warning signal; The misalignment and offset identification module acquires diastolic blood pressure, systolic blood pressure, arterial diameter and wall stress parameters, extracts peak positions, calculates relative offset differences, and generates offset sequence curves. When the fluctuation rate of the offset sequence curve exceeds a preset fluctuation density threshold, a set of cross-abnormal points is established. The gradient mutation determination module acquires the time series of systolic blood pressure, aortic diameter and heart rate, constructs the change rate trajectory vector and maps it to a unified time coordinate axis, calculates the time difference of local extreme points, and generates abnormal gradient combination sequences based on the time difference. The joint risk warning module constructs a set of misalignment matrix overlap density values ​​based on the electrolyte abnormality warning signal, the cross-abnormal point set, and the abnormal gradient combination sequence, and matches them with the risk pattern to generate aortic disease risk warning results.

2. The aortic disease risk early warning system based on big data according to claim 1, characterized in that, The two-factor mutation vector includes normalized spectral energy coordinates, ion concentration gradient dimension, and joint feature weight coefficients. The electrolyte abnormality warning signal includes Euclidean distance metric, lesion state classification label, and abnormality severity level. The cross-abnormality point set includes peak misalignment timestamp, abnormal parameter identifier, and interval fluctuation amplitude value. The abnormal gradient combination sequence includes rate mutation value, asynchronous time difference term, and gradient chain index. The aortic disease risk warning result includes model matching similarity score, risk evolution stage, and dominant triggering factor.

3. The aortic disease risk early warning system based on big data according to claim 1, characterized in that, The joint vector construction module includes: The frequency domain feature extraction submodule acquires the electrocardiogram signal for a specified time period, performs discrete Fourier transform on the acquired time domain waveform data, maps the time domain amplitude to the frequency domain space to extract the spectral energy distribution data in the range of 0.5Hz to 40Hz, performs integral processing on the energy spectral density of the high-frequency sub-interval, calculates the proportion of the integral value to the total energy of the entire frequency band, and generates the high-frequency energy proportion change rate based on the proportion fluctuation amplitude within the continuous time window. The ion concentration gradient calculation submodule obtains continuous blood potassium value measurement results, extracts the concentration values ​​and corresponding timestamps of adjacent sampling points in chronological order, performs a subtraction operation on the concentration values ​​of the previous and next time points to obtain the blood potassium concentration increment, calculates the time span between the corresponding sampling time points, performs a ratio operation, divides the increment value by the sampling interval duration to obtain the concentration change amplitude per unit time, and generates a gradient factor. The vector normalization splicing submodule calls the high-frequency band energy ratio change rate and gradient factor to calculate the numerical range of each feature column. The original value is subtracted from the minimum value in the column and then divided by the range to perform normalization mapping. According to the feature dimension alignment rule, the processed energy features and gradient features are sequentially spliced ​​to construct a two-factor mutation vector.

4. The aortic disease risk early warning system based on big data according to claim 3, characterized in that, The distance discrimination analysis module includes: The distance metric calculation submodule calls the two-factor mutation vector, accesses a pre-set sample library of approximate disease states, traverses the sample data in the library, calculates the numerical difference between the two-factor mutation vector and the sample in each feature dimension, performs square accumulation and arithmetic square root operations on the numerical difference, obtains the Euclidean space distance value, and generates a vector distance distribution set. The neighborhood instance filtering submodule obtains a preset distance radius threshold based on the vector distance distribution set, compares each Euclidean space distance value in the set with the distance radius threshold, filters the nearest neighbor vector objects whose distance is less than the distance radius threshold, extracts the object identifier and establishes an index association pointing to the original data, and generates an abnormal candidate instance list. The status determination and early warning submodule counts the cumulative number of nearest neighbor vector objects in the abnormal candidate instance list, and determines whether the number meets the minimum density requirement for lesion interval determination in combination with the time window. When it meets the requirement, it confirms that the aortic electrolyte state has entered the induced lesion interval and generates an electrolyte abnormality early warning signal.

5. The aortic disease risk early warning system based on big data according to claim 4, characterized in that, The process of setting the preset distance radius threshold specifically involves obtaining a historical distribution set of aortic standard sample data in a physiologically stable state; and calling statistical analysis functions to calculate the average intra-class spatial distance and standard deviation of intra-class spatial distance between sample points in the historical distribution set. Match the corresponding statistical multiple factor according to the preset risk sensitivity level; perform an addition operation on the product of the mean spatial distance within the class, the standard deviation of spatial distance within the class, and the statistical multiple factor to obtain the preset distance radius threshold; The process of determining whether the quantity meets the minimum density requirement for lesion interval determination by combining the time window is specifically as follows: within a preset sliding time window, frequency statistics are performed on the nearest neighbor vector objects included in the continuously generated list of abnormal candidate instances; the cumulative quantity obtained by statistics is compared with the minimum density requirement for lesion interval determination; when the cumulative quantity is greater than or equal to the minimum density requirement for lesion interval determination, it is determined that the minimum density requirement for lesion interval determination is met.

6. The aortic disease risk early warning system based on big data according to claim 4, characterized in that, The misalignment detection module includes: The parameter feature extraction submodule obtains parameters such as diastolic blood pressure, systolic blood pressure, arterial diameter, and wall stress. It performs extreme value search on the time-domain waveform of each parameter, locates local maxima within a continuous cardiac cycle, records the time coordinates, and generates a multidimensional peak position time series set. The offset sequence operation submodule calls the multidimensional peak position time series set, calculates the time difference between adjacent peak points to determine the relative offset interval length difference, introduces the standard cardiac cycle duration, blood pressure pulsatility coefficient, wall stress intensity factor, arterial diameter expansion index, atrioventricular synchronization deviation and blood flow velocity gradient modulus, calculates and obtains the offset sequence intensity value, arranges the continuously calculated intensity values ​​in time order, and generates the offset sequence curve. The abnormal pattern aggregation submodule obtains a preset physiological stable interval based on the offset sequence curve, compares the curve values ​​with the interval boundary point by point, counts the proportion of data points exceeding the boundary to calculate the fluctuation rate, compares the fluctuation rate with a preset fluctuation density threshold, and extracts the parameter combination pattern within the corresponding time period when it exceeds the fluctuation density threshold to establish a cross-abnormal point set.

7. The aortic disease risk early warning system based on big data according to claim 6, characterized in that, The specific formula for obtaining the offset sequence strength value is as follows: ； in, Represents the intensity value of the offset sequence. Represents the total number of sampling points. Representing the The difference in the relative offset interval length of each sampling point This represents the standard cardiac cycle duration. Represents the blood pressure pulsatility coefficient. Represents the wall stress intensity factor. Represents the base of the natural logarithm. Represents the arterial diameter dilation index. This represents the atrioventricular synchrony deviation. This represents the blood flow velocity gradient modulus.

8. The aortic disease risk early warning system based on big data according to claim 6, characterized in that, The gradient mutation determination module includes: The rate trajectory generation submodule acquires the time series of systolic blood pressure, aortic diameter and heart rate, performs discrete difference operation on each parameter sequence, calculates the ratio of the numerical increment between adjacent sampling points to the time step, constructs a first derivative sequence characterizing the dynamic change rate of the parameters, and combines the derivative sequences of each parameter by dimension to generate a change rate trajectory vector. The asynchronous time difference calculation submodule calls the change rate trajectory vector, uses linear interpolation logic to map each component to a unified time coordinate axis to achieve granular alignment, searches for local maxima of each component curve through a sliding window and extracts the corresponding timestamp, calculates the absolute value of the time interval between extreme points among multiple parameter types, and generates local extreme asynchronous time difference. The gradient sequence determination submodule obtains a preset abnormal asynchronous threshold range based on the local extreme asynchronous time difference, filters time difference data that fall into the abnormal asynchronous threshold range, detects whether the time difference values ​​in multiple consecutive cardiac cycles show a monotonically increasing or decreasing evolution trend, extracts time difference data that meet the trend characteristics and their corresponding gradient indices, and generates an abnormal gradient combination sequence.

9. The aortic disease risk early warning system based on big data according to claim 8, characterized in that, The joint risk warning module includes: The matrix density construction submodule calls the electrolyte abnormality warning signal, the cross-abnormal point set and the abnormal gradient combination sequence, extracts the time index and intensity parameters, maps them to a multi-dimensional feature space grid with a preset resolution, traverses all grid cells in the multi-dimensional space to count the cumulative number of abnormal feature points falling into them, calculates the ratio of the cumulative number to the grid space capacity, and generates a set of misaligned matrix overlap density values. The pattern matching calculation submodule obtains a preset high-risk pattern density template based on the set of overlapping density values ​​of the misaligned matrix, expands the density set and template matrix into a one-dimensional numerical vector, performs multiplication and accumulation operations on the two vectors respectively to obtain the inner product value, calculates the Euclidean norm product of each vector, performs division operations to divide the inner product value by the Euclidean norm product, and generates a model matching similarity score. The risk warning generation submodule matches the similarity score of the model, obtains a preset warning trigger threshold, compares the score with the warning trigger threshold, filters calculation instances whose scores exceed the warning trigger threshold, retrieves the corresponding lesion evolution stage description information according to the numerical range to which the score belongs, aggregates the risk level code and main triggering factors at the current moment, and generates aortic disease risk warning results.

Images