Heart failure risk assessment method and system based on AI

Abstract

The invention provides an AI-based heart failure risk assessment method and system, and relates to the technical field of artificial intelligence, and the method comprises the steps: recognizing a mutation point of a physiological index time sequence through a recursive segmentation algorithm, extracting a mutation feature matrix, carrying out the time sequence mapping, recognizing a periodic mutation and gradual change mode, and converting into a risk score to generate a risk accumulation curve. Calculating index conduction time delay, drawing a risk conduction path diagram, marking an intervention time window, and finally determining a heart failure risk level and generating an evaluation report. According to the invention, early accurate recognition of the heart failure risk can be realized, and the best opportunity is provided for clinical intervention.

Description

Technical Field

[0001] The present invention relates to artificial intelligence technology, and in particular to a heart failure risk assessment method and system based on AI. Background Art

[0002] Heart failure is a common cardiovascular disease. Due to the insufficient heart pumping function, the body cannot meet the metabolic needs, which is characterized by high incidence, high mortality and high readmission rate. With the aggravation of population aging, the number of heart failure patients increases year by year, bringing a heavy burden to the medical system. Early risk assessment is crucial for the prevention and intervention of heart failure, which can effectively reduce the incidence of heart failure and the risk of its complications.

[0003] Traditional heart failure risk assessment mainly relies on clinicians' empirical judgment based on patients' medical history, physical examination and laboratory tests. With the development of medical technology, wearable devices and remote monitoring systems can continuously collect a variety of physiological indicators of patients, providing a large amount of real-time data for the dynamic assessment of heart failure risk. The rise of artificial intelligence technology provides new ideas for processing and analyzing these complex medical data.

[0004] The existing heart failure risk assessment technologies have the following deficiencies: Most assessment methods handle time series data relatively simply, usually only focusing on outliers at a single time point or static statistical features, and unable to capture the complex change patterns and mutation characteristics of physiological indicators in the time dimension, resulting in insufficient sensitivity to early minor changes in the disease. Existing assessment models often consider various physiological indicators in isolation, lacking in-depth analysis of the mutual relationships between multi-dimensional indicators, and unable to effectively identify the conduction relationships and time series dependencies between indicators, thus missing important pathophysiological mechanism clues in the development process of heart failure. The existing technologies are relatively simple in presenting the risk assessment results, usually only giving the risk level or probability value, lacking the precise identification of the intervention time window and the detailed description of specific abnormal indicators, and it is difficult to provide targeted intervention suggestions for clinicians. Summary of the Invention

[0005] An embodiment of the present invention provides a heart failure risk assessment method and system based on AI, which can solve the problems in the existing technology.

[0006] In the first aspect of the embodiment of the present invention, a heart failure risk assessment method based on AI is provided, including: Collect the physiological indicators of the object to be evaluated to construct a multi-dimensional time series data stream, calculate the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm, determine the time points with the difference degree greater than the preset difference threshold as mutation points, extract the waveform features of the data segments before and after the mutation points, and generate a mutation feature matrix; Perform a temporal mapping on the mutation feature matrix, obtain the time interval sequence and amplitude sequence of the mutation features, identify the periodic mutation patterns and gradual change patterns, convert the duration and occurrence frequency of the periodic mutation patterns and gradual change patterns into risk scores, and draw a risk accumulation curve; Calculate the change temporal relationship between physiological indicators according to the risk accumulation curve, convert the change temporal relationship into the index conduction delay, draw a risk conduction path diagram based on the index conduction delay, and mark the time interval with the largest change slope in the risk conduction path diagram as the intervention time window; Based on the time range of the intervention time window and the change trend of the risk accumulation curve, determine the heart failure risk level, and generate a risk assessment report including the risk level and abnormal description of physiological indicators.

[0007] In an alternative embodiment, Collect the physiological indicators of the object to be evaluated and construct them into a multi-dimensional time series data stream. Calculate the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm, and determine the mutation points where the difference degree is greater than the preset difference threshold, including: Collect the physiological indicators of the object to be evaluated, and the physiological indicators include cardiovascular data, respiratory data, and blood oxygen data; resample the physiological indicators through cubic spline interpolation to construct a multi-dimensional time series data stream, and construct the multi-dimensional time series data stream into a state matrix of time-index dimension; Initial divide the state matrix from the midpoint position to obtain two sub-data segments, move the division position point by point along the time axis to calculate the Mahalanobis distance of the two-side data segments to obtain a distance sequence and perform eigenvalue decomposition to obtain the principal component weight vector, and use the principal component weight vector to weight the distance sequence to obtain the sub-data segment difference degree sequence; Statistically analyze the probability distribution characteristics of the sub-data segment difference degree sequence to obtain a cumulative distribution function, determine the difference threshold on the cumulative distribution function according to the preset false alarm rate, mark the time points greater than the difference threshold as initial mutation points, construct an observation window centered on the initial mutation points, calculate the mutual information between the signals of each dimension within the observation window to obtain a mutual information matrix, and use the sum of the non-diagonal elements of the mutual information matrix as the coupling metric value; Calculate the comprehensive difference degree based on the coupling metric value and the sub-data segment difference degree sequence, screen the initial mutation points according to the comprehensive difference degree, and determine the time points with the comprehensive difference degree greater than the difference threshold as the final mutation points.

[0008] In an alternative embodiment, Extract the waveform features of the data segments before and after the mutation points, and generate a mutation feature matrix including: Calculate the volatility characteristics of the data segments before and after the mutation point based on the mutation point position to obtain a volatility sequence, and perform segmented cumulative calculation to obtain a volatility trend feature. Determine an adaptive window length based on the volatility trend feature, and use the adaptive window length to determine the waveform feature extraction intervals before and after the mutation point; Within the waveform feature extraction intervals, perform time-frequency transformation on the data segments before and after the mutation point respectively to obtain time-frequency spectra, calculate the energy distribution of the time-frequency spectra to obtain a time-frequency energy sequence, and perform weighted fusion on the time-frequency energy sequence in combination with the volatility sequence to obtain a fusion feature sequence. Calculate the multi-scale entropy value of the fusion feature sequence to obtain a complexity feature; Combine the volatility sequence, the time-frequency energy sequence, and the complexity feature in chronological order to obtain a feature vector. Construct an initial feature matrix based on the feature vector, calculate the discriminant contribution of each feature in the initial feature matrix to obtain feature weights, and use the feature weights to perform weighted reconstruction on the initial feature matrix to generate a mutation feature matrix.

[0009] In an alternative embodiment, Perform chronological mapping on the mutation feature matrix, obtain the time interval sequence and amplitude sequence of the mutation features, identify periodic mutation patterns and gradual change patterns, and convert the duration and occurrence frequency of the periodic mutation patterns and gradual change patterns into risk scores. Draw a risk accumulation curve including: Perform chronological mapping on the mutation feature matrix to obtain a time interval sequence and an amplitude sequence. Construct a mutation topology network based on the time interval sequence and the amplitude sequence through an adaptive weight function, where the nodes represent mutation points, and the connection weights between the nodes are jointly determined by the time interval and the amplitude; Calculate the local entropy of the nodes at different scales of the mutation topology network, determine the optimal scale set according to the change characteristics of the local entropy, and fuse the local entropy under the optimal scale set to obtain an enhanced topological entropy sequence; Dynamically adjust the calculation window size based on the change rate of the enhanced topological entropy sequence. Calculate the autocorrelation function and power spectral density of the enhanced topological entropy sequence within the calculation window, extract periodic mutation features, determine the first duration and the first occurrence frequency of the periodic mutation pattern according to the periodic mutation features, perform piecewise polynomial fitting on the enhanced topological entropy sequence to identify the gradual change interval, and determine the second duration and the second occurrence frequency of the gradual change pattern according to the gradual change interval; Convert the first duration and the first occurrence frequency into a first risk score through a first mapping function, convert the second duration and the second occurrence frequency into a second risk score through a second mapping function, obtain a risk score according to the dynamic weighted combination of the first risk score and the second risk score, and perform time integration on the risk score to obtain a risk accumulation curve.

[0010] In an alternative embodiment, Calculate the local entropy of nodes at different scales of the mutant topological network, determine the optimal scale set according to the change characteristics of the local entropy, and fuse the local entropy under the optimal scale set to obtain an enhanced topological entropy sequence, including: Construct a node neighborhood matrix in the mutant topological network, where the node neighborhood matrix represents the set of nodes reachable by each node within a specified number of steps. Based on the node neighborhood matrix, construct a node feature vector including degree centrality and clustering coefficient, calculate the node structure similarity, and adaptively reconstruct the network connection using the node structure similarity; Calculate the cumulative weight between nodes in the reconstructed network, where the cumulative weight is obtained by accumulating recursively at different numbers of steps. Based on the cumulative weight, calculate the transition probability between nodes, and adaptively non-linearly adjust the transition probability according to the local density distribution of the network to obtain a corrected transition probability; Calculate the local entropy of nodes at different scales based on the corrected transition probability to obtain a local entropy sequence, calculate the multi-order difference characteristics and community evolution characteristics of the local entropy sequence, and construct a scale importance index; Determine the optimal scale set according to the scale importance index, calculate the information redundancy of each scale in the optimal scale set based on the complementary measure, use the normalized reciprocal of the information redundancy as the fusion weight, and perform a weighted combination of the local entropy sequence under the optimal scale set using the fusion weight to obtain an enhanced topological entropy sequence.

[0011] In an alternative embodiment, Calculate the change time sequence relationship between physiological indicators according to the risk accumulation curve, convert the change time sequence relationship into an index conduction delay, draw a risk conduction path map based on the index conduction delay, and mark the time interval with the largest change slope in the risk conduction path map as the intervention time window, including: Extract the local features of the risk accumulation curve at multiple time window scales, calculate the distance matrix between physiological indicators according to the local features, and calculate the change time sequence relationship between the physiological indicators using a time warping algorithm with attention weights based on the distance matrix; Construct a hierarchical coupling network between the physiological indicators, recursively calculate the direct coupling coefficient and the indirect coupling coefficient in the hierarchical coupling network, and perform a weighted correction on the change time sequence relationship to obtain an index conduction delay; Construct a risk conduction path map with the physiological indicators as nodes and the index conduction delay as edge weights, calculate the dynamic score of the nodes in the risk conduction path map using a time-varying node importance algorithm, and determine the conduction path according to the dynamic score and the index conduction delay; An adaptive segmented planning algorithm is used on the conduction path to search for the interval with the maximum slope to obtain a candidate time window, the information entropy of the slope sequence within the candidate time window is calculated to obtain the window confidence level. When the window confidence level is lower than a preset confidence threshold, the interval range is expanded along the conduction path to re-search for the interval with the maximum slope, and the interval with the maximum slope that satisfies the window confidence level is used as the final intervention time window.

[0012] In an alternative embodiment, Construct a hierarchical coupling network among the physiological indicators. Recursively calculate the direct coupling coefficient and the indirect coupling coefficient in the hierarchical coupling network, and perform weighted correction on the change time sequence relationship to obtain the index conduction delay, including: Perform sliding window segmentation on the change time sequence relationship between physiological indicators to obtain local time sequence segments. Use the dynamic time warping algorithm to calculate the matching degree of the local time sequence segments to obtain a time sequence similarity matrix. Perform multi-scale wavelet transform on the local time sequence segments for frequency domain decomposition, calculate the phase synchrony of different frequency components to obtain a frequency coupling degree matrix, and perform adaptive weighted fusion on the time sequence similarity matrix and the frequency coupling degree matrix to obtain the coupling strength between indicators; Construct an initial hierarchical coupling network based on the coupling strength between indicators, calculate the conditional mutual information between network nodes, perform a causal entropy significance test based on the conditional mutual information to screen for stable coupling edges, use the stable coupling edges to update the initial hierarchical coupling network to obtain an optimized hierarchical coupling network, and extract the coupling strength of the network adjacent edges in the optimized hierarchical coupling network to obtain the direct coupling coefficient; Perform a depth-first search on the optimized hierarchical coupling network to identify candidate conduction links between physiological indicators, calculate the information entropy of the candidate conduction links, set an effective threshold based on the information entropy to screen for effective conduction links, perform recursive calculation on the effective conduction links, and use the ratio of the cumulative effect of the coupling strength of each edge on the link to the path length as the indirect coupling coefficient; Calculate the comprehensive coupling coefficient based on the direct coupling coefficient and the indirect coupling coefficient, use the comprehensive coupling coefficient to perform non-linear weighted correction on the change time sequence relationship, normalize the corrected change time sequence relationship and convert it into time units to obtain the index conduction delay.

[0013] In the second aspect of the embodiments of the present invention, an AI-based heart failure risk assessment system is provided, including: A first unit for collecting physiological indicators of an object to be evaluated to construct a multi-dimensional time series data stream, calculating the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm, determining the time points with a difference degree greater than a preset difference threshold as mutation points, extracting the waveform features of the data segments before and after the mutation points, and generating a mutation feature matrix; A second unit, configured to perform a temporal mapping on the mutation feature matrix, obtain a time interval sequence and an amplitude sequence of the mutation features, identify periodic mutation patterns and gradual change patterns, convert the duration and occurrence frequency of the periodic mutation patterns and the gradual change patterns into risk scores, and draw a risk accumulation curve; A third unit, configured to calculate the changing temporal relationship between physiological indicators according to the risk accumulation curve, convert the changing temporal relationship into an indicator conduction delay, draw a risk conduction path diagram based on the indicator conduction delay, and mark, in the risk conduction path diagram, the time interval with the largest change slope as the intervention time window; A fourth unit, configured to determine a heart failure risk level based on the time range of the intervention time window and the change trend of the risk accumulation curve, and generate a risk assessment report including the risk level and a description of abnormal physiological indicators.

[0014] In a third aspect of the embodiments of the present invention, there is provided an electronic device, including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0015] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0016] In this embodiment, the AI-based heart failure risk assessment method identifies mutation points in multi-dimensional time series data through a recursive partitioning algorithm and extracts a mutation feature matrix, which can accurately capture the critical moments of cardiac function changes, improving the accuracy and sensitivity of abnormal pattern recognition. Through temporal mapping and pattern recognition of mutation features, this method can effectively distinguish periodic mutations and gradual change patterns, convert complex physiological signal changes into quantifiable risk scores, realize dynamic assessment and visual display of heart failure risks, and facilitate medical staff to intuitively understand the patient's condition. The concept of indicator conduction delay is innovatively introduced and a risk conduction path diagram is drawn, which can not only reveal the internal relationship between various physiological indicators, but also accurately mark the optimal intervention time window, providing a scientific basis for clinical intervention and effectively avoiding the problem of poor treatment effects caused by improper timing of intervention in traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a schematic flowchart of the AI-based heart failure risk assessment method according to the embodiments of the present invention; Figure 2 is a performance comparison diagram of mutation point detection in multi-dimensional physiological signals according to the embodiments of the present invention; Figure 3Performance comparison chart of the risk assessment and early warning system according to the embodiments of the present invention; Figure 4 Delay calculation accuracy comparison chart according to the embodiments of the present invention. Specific implementation manners

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0020] Figure 1 Schematic flowchart of the AI-based heart failure risk assessment method according to the embodiments of the present invention, as Figure 1 shown, the method includes: Collect physiological indicators of the object to be evaluated to construct a multi-dimensional time series data stream, calculate the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm, determine the time points with a difference degree greater than a preset difference threshold as mutation points, extract the waveform features of the data segments before and after the mutation points, and generate a mutation feature matrix; Perform a time series mapping on the mutation feature matrix, obtain the time interval sequence and amplitude sequence of the mutation features, identify the periodic mutation pattern and the gradual change pattern, convert the duration and occurrence frequency of the periodic mutation pattern and the gradual change pattern into risk scores, and draw a risk accumulation curve; Calculate the changing time series relationship between physiological indicators according to the risk accumulation curve, convert the changing time series relationship into an index conduction delay, draw a risk conduction path diagram based on the index conduction delay, and mark the time interval with the largest change slope in the risk conduction path diagram as the intervention time window; Based on the time range of the intervention time window and the changing trend of the risk accumulation curve, determine the heart failure risk level, and generate a risk assessment report including the risk level and abnormal description of physiological indicators.

[0021] In an alternative implementation manner, collecting physiological indicators of the object to be evaluated to construct a multi-dimensional time series data stream, calculating the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm, and determining the time points with a difference degree greater than a preset difference threshold as mutation points includes: Collect the physiological indicators of the object to be evaluated, where the physiological indicators include cardiovascular data, respiratory data, and blood oxygen data; resample the physiological indicators through cubic spline interpolation to construct a multi-dimensional time series data stream, and construct the multi-dimensional time series data stream into a state matrix of time-index dimension; Initial divide the state matrix from the midpoint position to obtain two sub-data segments, calculate the Mahalanobis distance of the two-side data segments by moving the division position point by point along the time axis to obtain a distance sequence and perform eigenvalue decomposition to obtain the principal component weight vector, and use the principal component weight vector to weight the distance sequence to obtain the sub-data segment difference degree sequence; Statistically analyze the probability distribution characteristics of the sub-data segment difference degree sequence to obtain a cumulative distribution function, determine a difference threshold on the cumulative distribution function according to a preset false alarm rate, mark the time points greater than the difference threshold as initial mutation points, construct an observation window centered on the initial mutation points, calculate the mutual information between the signals of each dimension within the observation window to obtain a mutual information matrix, and use the sum of the non-diagonal elements of the mutual information matrix as the coupling metric value; Calculate the comprehensive difference degree based on the coupling metric value and the sub-data segment difference degree sequence, screen the initial mutation points according to the comprehensive difference degree, and determine the time points with the comprehensive difference degree greater than the difference threshold as the final mutation points.

[0022] In order to accurately monitor the physiological state of the object to be evaluated, this embodiment provides a method for detecting mutation points of physiological indicators. This method calculates the difference degree between adjacent data segments in a multi-dimensional time series data stream through a recursive segmentation algorithm to realize the identification and monitoring of physiological state mutations. First, collect the physiological indicators of the object to be evaluated, including cardiovascular data, respiratory data, and blood oxygen data. The cardiovascular data may include data such as heart rate and blood pressure; the respiratory data may include data such as respiratory rate and respiratory depth; the blood oxygen data may include data such as blood oxygen saturation and blood oxygen fluctuation range. For example, for continuous monitoring of a certain patient, heart rate data (fluctuating between 60 and 80 times per minute), respiratory rate data (fluctuating between 14 and 18 times per minute), and blood oxygen saturation data (fluctuating within the range of 95%-99%) are collected.

[0023] The collected physiological index data is resampled by cubic spline interpolation to construct a multi-dimensional time series data stream. Since the sampling frequencies of various physiological indexes may be inconsistent, for example, the heart rate data is sampled once every 1 second, while the respiratory data is sampled once every 2 seconds. To ensure the time consistency of the data, the cubic spline interpolation method is used to resample the original data so that all indexes have the same sampling time points. After resampling, the multi-dimensional time series data stream is constructed into a state matrix of time-index dimension, where the rows of the matrix represent time points and the columns represent different physiological indexes. Then, for the constructed state matrix, an initial division is made from the midpoint position to obtain two sub-data segments. For example, for a state matrix containing 1000 time points, the initial division is made from the 500th time point to obtain two sub-data segments each containing 500 time points. Subsequently, the division position is moved point by point along the time axis, and the Mahalanobis distances of the two data segments on both sides are calculated respectively. The specific operation is as follows: Assume that the time point t is the current division point, calculate the Mahalanobis distance between the data segment from time point 1 to t and the data segment from time point t + 1 to the end, and obtain a distance value; then move the division point to t + 1 and repeat the above calculation process to finally obtain a distance sequence. For example, for a state matrix with a length of 1000, 998 Mahalanobis distance values (from position 2 to position 999) will be obtained.

[0024] Perform eigenvalue decomposition on the obtained distance sequence to obtain the principal component weight vector. Eigenvalue decomposition can reveal the main change patterns in the distance sequence and extract the principal component weight vector. Use the principal component weight vector to weight the distance sequence to obtain the sub-data segment difference degree sequence. The weight vector reflects the contribution degree of each time point to the overall difference, and the weighted difference degree sequence can more accurately express the degree of state change.

[0025] Statistically analyze the probability distribution characteristics of the sub-data segment difference degree sequence to obtain the cumulative distribution function. According to the preset false alarm rate (such as 0.05), determine the difference threshold on the cumulative distribution function. For example, if the false alarm rate is set to 0.05, the difference threshold is the 95% quantile of the difference degree sequence. Mark the time points greater than this threshold as initial mutation points. For each initial mutation point, construct an observation window centered on it. The size of the observation window can be set according to the actual application scenario. For example, a window of 21 time points with 10 time points before and after each is used. Calculate the mutual information between the signals of each dimension within the observation window to obtain the mutual information matrix. Each element in the mutual information matrix represents the mutual information value between two-dimensional signals, reflecting the correlation between the signals. For example, for data containing three dimensions of heart rate, respiratory rate, and blood oxygen saturation, the mutual information matrix is a 3×3 matrix.

[0026] Take the sum of the non - diagonal elements of the mutual information matrix as the coupling metric value. The coupling metric value reflects the coupling degree among the signals of each dimension within the observation window. The larger the value, the more coordinated the changes of the signals of each dimension. Combine the coupling metric value with the sub - data - segment difference degree sequence to calculate the comprehensive difference degree. The comprehensive difference degree can be obtained by weighted average. For example, set the weight of the coupling metric value to 0.4 and the weight of the sub - data - segment difference degree to 0.6, and calculate the comprehensive difference degree through weighted calculation.

[0027] Screen the initial mutation points according to the comprehensive difference degree, and determine the time points with the comprehensive difference degree greater than the difference threshold as the final mutation points. For example, in the data analysis of a certain patient, among 1000 time points, 15 mutation points are initially detected. After screening by the comprehensive difference degree, 8 time points are determined as the final mutation points. These time points correspond to significant changes in the patient's physiological state, such as rest after strenuous exercise, mood fluctuations, or the onset of drug effects, etc.

[0028] In this embodiment, through cubic spline interpolation resampling and the construction of multi - dimensional time - series data streams, the scheme can effectively handle the time - series property and complexity of physiological data, ensuring the smoothness and continuity of the data. On this basis, by using the method combining Mahalanobis distance and principal component analysis, the difference degree between each time period can be quantified, and potential mutation points can be accurately identified. In addition, by calculating the mutual information between signals and combining the coupling metric value, the evaluation accuracy of the correlation between signals of different dimensions can be improved, ensuring the more accurate and reliable identification of mutation points. Finally, by screening mutation points through the comprehensive difference degree, the scheme can effectively reduce the false - alarm rate and achieve efficient and accurate abnormal detection of physiological data.

[0029] Figure 2 This is a performance comparison chart of mutation point detection in multi - dimensional physiological signals in the embodiment of the present invention, as Figure 2As shown, this figure shows the performance comparison of different mutation point detection methods. The horizontal axis represents the false alarm rate, and the vertical axis represents the detection rate. The closer the curve is to the upper left corner, the better the performance. This technical solution shows the highest detection rate at each false alarm rate level. Especially when the false alarm rate is 0.2, the detection rate reaches 0.85, which is marked as the optimal operating point. When the false alarm rate is 0.6, the detection rate of this technical solution is as high as 96.3%, while the detection rates of the traditional spectral analysis method and the traditional threshold method are only 82.7% and 89.1% respectively under the same conditions. The shaded area in the figure intuitively shows the performance improvement of this technical solution compared with the traditional method. This significant advantage mainly comes from the innovative algorithm of combining Mahalanobis distance with principal component weighting in this technical solution, and the method of introducing the mutual information matrix to calculate the coupling degree to further screen mutation points. Compared with the traditional spectral analysis method (such as Fourier transform spectral analysis) and the simple threshold method (such as fixed threshold or adaptive threshold method), this solution has achieved a qualitative leap in the balance between false alarm rate control and detection rate, especially the detection ability under low false alarm rate conditions is more prominent, which has important clinical value for the accurate identification of key mutation points in physiological signals.

[0030] In an alternative embodiment, waveform features of the data segments before and after the mutation point are extracted, and generating a mutation feature matrix includes: Calculating the volatility sequence of the data segments before and after the mutation point according to the mutation point position to obtain the volatility sequence, and performing segmented cumulative calculation to obtain the fluctuation trend feature. Based on the fluctuation trend feature, an adaptive window length is determined, and the waveform feature extraction interval before and after the mutation point is determined by using the adaptive window length; Within the waveform feature extraction interval, performing time-frequency transformation on the data segments before and after the mutation point respectively to obtain the time-frequency spectrum, calculating the energy distribution of the time-frequency spectrum to obtain the time-frequency energy sequence, and performing weighted fusion on the time-frequency energy sequence in combination with the volatility sequence to obtain the fusion feature sequence, and calculating the multi-scale entropy value of the fusion feature sequence to obtain the complexity feature; Combining the volatility sequence, the time-frequency energy sequence, and the complexity feature in chronological order to obtain a feature vector, constructing an initial feature matrix according to the feature vector, calculating the discriminant contribution degree of each feature in the initial feature matrix to obtain the feature weight, and using the feature weight to perform weighted reconstruction on the initial feature matrix to generate a mutation feature matrix.

[0031] This embodiment provides a method for extracting waveform features of data segments before and after mutation points and generating a mutation feature matrix. This method mainly includes three parts: determining the adaptive window length, extracting waveform features, and constructing the feature matrix. Specifically, first, obtain the original data sequence containing the mutation point, assumed to be a sequence X of length N, and the mutation point position is p. Taking the voltage waveform in the power system as an example, voltage data with a sampling rate of 10 kHz within 10 seconds can be obtained, with a total length of 100,000 sampling points, and the mutation point is located at the 45,000th sampling point.

[0032] Calculate the volatility sequences of the data before and after the mutation point. Take M data points before and after the mutation point, which are sequences X(p - M:p - 1) and X(p:p + M - 1) respectively, where M can be taken as N / 10. For each subsequence, generate volatility sequences V1 and V2 by calculating the absolute value of the difference between adjacent data points and normalizing. For example, for voltage data, when M = 10,000, volatility sequences of 10,000 points before and after the mutation point are obtained.

[0033] Perform segmented cumulative calculation on the volatility sequences to obtain the fluctuation trend features. Divide the volatility sequences V1 and V2 into K equal - length segments (such as K = 10), calculate the cumulative sum of the volatility within each segment, and form the fluctuation trend sequences T1 and T2. For example, divide the 10,000 - point volatility sequence into 10 segments, each segment has 1,000 points, calculate the cumulative sum of the volatility of each segment, and obtain a fluctuation trend sequence of length 10.

[0034] Determine the adaptive window length based on the fluctuation trend features. Calculate the change rates of the fluctuation trend sequences T1 and T2, take the L segments (such as L = 3) with the largest change rates, and accumulate the lengths of these segments as the adaptive window length W. In an actual case, if the change rates of the 2nd, 5th, and 8th segments are the largest, and the length of each segment is 1,000 points, then the adaptive window length W is 3,000 points.

[0035] Extract waveform features within the determined adaptive window length. First, determine the waveform feature extraction interval. Taking the mutation point p as the center, take W points forward and W points backward to form the feature extraction interval [p - W, p + W]. For example, when W = 3,000, the feature extraction interval is [42,000, 48,000], which contains a total of 6,000 data points. Perform time - frequency transformation on the data segments before and after the mutation point. Perform short - time Fourier transform on the data in the intervals [p - W, p - 1] and [p, p + W] respectively, set the window length to W / 10, and the window overlap rate to 50%, to obtain two time - frequency spectra S1 and S2. For example, when the window length is 300 points, the obtained time - frequency spectrum has 19 time points in the time dimension and 150 frequency points in the frequency dimension.

[0036] Calculate the energy distribution of the time-frequency spectrum. For each time-frequency spectrum S1 and S2, calculate the energy sum of each frequency band along the frequency dimension to form time-frequency energy sequences E1 and E2. These two sequences reflect the frequency band energy distribution characteristics of the data at different time points. For example, an energy sequence of length 19 is obtained, and each value represents the total energy of all frequency bands at the corresponding time point. Combine the time-frequency energy sequences by weighting with the volatility sequences. Resample the volatility sequences V1 and V2 to the same length as E1 and E2, perform point-by-point multiplication operations to obtain the fused feature sequences F1 and F2. For example, resample a volatility sequence of 3,000 points to 19 points and multiply it with the energy sequence at the corresponding positions to obtain the fused feature sequence.

[0037] Calculate the multi-scale entropy values of the fused feature sequences. Calculate the sample entropy for F1 and F2 respectively, set the scale factor from 1 to 5, the embedding dimension to 2, and the similarity tolerance to 0.15 times the sequence standard deviation to obtain the multi-scale entropy value sequences H1 and H2 as complexity features. For example, calculate the entropy values of 5 scales for each fused sequence to obtain a complexity feature of length 5.

[0038] Combine the volatility sequences, time-frequency energy sequences, and complexity features in chronological order to form feature vectors. For the data segments before and after the mutation points, combine V1, E1, H1 and V2, E2, H2 respectively to obtain the feature vectors FV1 and FV2. For example, connect a volatility sequence of length 19, an energy sequence of length 19, and a complexity feature of length 5 to obtain a feature vector of length 43. Combine the feature vectors FV1 and FV2 to construct the initial feature matrix FM. Each row of the matrix corresponds to a mutation sample, and each column corresponds to a feature dimension. For example, for 100 mutation samples, the size of the obtained initial feature matrix is 100×86 (43×2 feature dimensions for each sample).

[0039] Calculate the discriminant contribution of each feature in the initial feature matrix. Use the information gain method to evaluate the contribution of each feature column to distinguishing different types of mutations, and generate the feature weight vector W. For example, for 86-dimensional features, 86 corresponding weight values are calculated, and the larger the value, the greater the contribution of the feature to the discrimination of mutation types. Reconstruct the initial feature matrix by weighting using the feature weights. Perform a point-by-point multiplication operation on the feature weight vector W and each column of the initial feature matrix FM to obtain the weighted mutation feature matrix WFM. For example, multiply the feature values of the i-th column by the corresponding weight W[i] to generate the reconstructed feature matrix.

[0040] In this embodiment, through the analysis of fluctuation characteristics and the determination of an adaptive window for fluctuation trend characteristics, the dynamic changes of waveforms can be effectively identified, and the extraction interval of waveform characteristics can be accurately defined. The application of time-frequency transformation can capture the changes of data in different frequency bands. The time-frequency energy sequence and the volatility sequence are weighted and fused, making the feature extraction more refined, so as to more accurately reflect the essential characteristics of the data. The calculation of multi-scale entropy values further enhances the ability to capture the changes in complexity, making the identification of mutation points more sensitive. Through the construction and weighted reconstruction of feature vectors, the scheme can optimize the discrimination ability of features, improve the sensitivity and accuracy of the model to mutation points, and thus effectively improve the accuracy and reliability of mutation detection.

[0041] In an alternative embodiment, a temporal mapping is performed on the mutation feature matrix to obtain a time interval sequence and an amplitude sequence of the mutation features, periodic mutation patterns and gradual change patterns are identified, and the duration and occurrence frequency of the periodic mutation patterns and gradual change patterns are converted into risk scores, and a risk accumulation curve is plotted, including: A temporal mapping is performed on the mutation feature matrix to obtain a time interval sequence and an amplitude sequence. Based on the time interval sequence and the amplitude sequence, a mutation topology network is constructed through an adaptive weight function, where nodes represent mutation points, and the connection weights between nodes are jointly determined by the time interval and the amplitude; The local entropy of nodes is calculated at different scales of the mutation topology network, and an optimal scale set is determined according to the change characteristics of the local entropy. The local entropy under the optimal scale set is fused to obtain an enhanced topological entropy sequence; Based on the change rate of the enhanced topological entropy sequence, the calculation window size is dynamically adjusted. The autocorrelation function and power spectral density of the enhanced topological entropy sequence are calculated within the calculation window, and periodic mutation features are extracted. According to the periodic mutation features, the first duration and the first occurrence frequency of the periodic mutation pattern are determined. The enhanced topological entropy sequence is segmented and polynomially fitted to identify the gradual change interval, and the second duration and the second occurrence frequency of the gradual change pattern are determined according to the gradual change interval; The first duration and the first occurrence frequency are converted into a first risk score through a first mapping function, the second duration and the second occurrence frequency are converted into a second risk score through a second mapping function, a risk score is obtained according to the dynamic weighted combination of the first risk score and the second risk score, and the risk score is time-integrated to obtain a risk accumulation curve.

[0042] This embodiment focuses on the analysis of mutation features in the heart failure risk assessment process, and realizes the identification and risk quantification of mutation patterns by constructing a mutation topology network. First, a temporal mapping is performed on the mutation feature matrix to obtain a time interval sequence and an amplitude sequence of the mutation features. The time interval sequence represents the time difference between adjacent mutation points, and the amplitude sequence represents the change amplitude of the physiological index corresponding to each mutation point.

[0043] When constructing a mutation topology network based on the time interval sequence and amplitude sequence, an adaptive weight function is used to determine the connection weights between nodes in the network. Specifically, for the connection weights between any two mutation points, they are obtained by the weighted combination of the normalized value of the time interval and the normalized value of the amplitude difference. For example, for the mutation points of the heart rate index, when the time interval between two mutation points is ten minutes and the amplitude difference is twenty beats per minute, the time interval can be normalized to the range of zero to one, and the amplitude difference can also be normalized to the same range. Then, the two normalized values are combined through an adaptive weight coefficient to obtain the connection weights between these two mutation points.

[0044] In the constructed mutation topology network, the local entropy of nodes is calculated by setting different observation scales. The local entropy reflects the aggregation degree and correlation strength of mutation points at different time scales. By analyzing the trend of the local entropy changing with the scale, the optimal scale set that can best reflect the mutation characteristics can be determined. For example, for the mutation points of the blood pressure index, the local entropy changes may be most significant at three time scales: five minutes, thirty minutes, and two hours, and these three scales constitute the optimal scale set.

[0045] Multiple local entropy sequences under the optimal scale set are weighted and fused to obtain an enhanced topological entropy sequence. The weight coefficients in the fusion process can be dynamically adjusted according to the significance of the local entropy at each scale. For example, for the respiratory rate index, if the local entropy fluctuates most violently at the five-minute scale, a larger weight coefficient is assigned to this scale.

[0046] Based on the change characteristics of the enhanced topological entropy sequence, the calculation window size is dynamically adjusted. When the sequence changes violently, a smaller calculation window is used to capture the rapid change characteristics; when the sequence is relatively stable, a larger calculation window is used to extract the long-term change trend. Within the determined calculation window, the autocorrelation function and power spectral density of the enhanced topological entropy sequence are calculated respectively.

[0047] By analyzing the periodicity of the autocorrelation function and the peak characteristics of the power spectral density, periodic mutation characteristics are extracted. For a mutation sequence with significant periodicity, the period length can be determined by the position of the first non-zero peak of the autocorrelation function, and the mutation intensity can be determined by the main frequency components of the power spectral density. For example, when analyzing the blood oxygen saturation index, if it is found that an obvious periodic mutation occurs every four hours and this periodic pattern repeats six times within twenty-four hours, the duration of the periodic mutation pattern can be recorded as twenty-four hours and the occurrence frequency as six times.

[0048] Perform piecewise polynomial fitting on the enhanced topological entropy sequence, and identify the gradual change intervals by analyzing the slope changes of the fitting curves. Calculate the average slope and duration of the fitting curves within each gradual change interval, and count the occurrence times of the gradual change intervals. For example, for the cardiac ejection fraction index, if a gradual change process with a duration of eight hours and an average decline rate of 0.5% per hour is detected, and similar gradual change processes occur three times during the observation period, then record the duration of the gradual change pattern as 24 hours and the occurrence frequency as three times.

[0049] Input the duration and occurrence frequency of the periodic mutation pattern into the first mapping function to obtain the risk score corresponding to the periodic mutation. The first mapping function adopts a piecewise continuous non-linear function form, with different mapping slopes in different duration and frequency intervals. For example, when the duration of the periodic mutation exceeds 24 hours or the frequency exceeds eight times per day, the slope of the mapping function will increase significantly, generating a higher risk score. Similarly, input the duration and occurrence frequency of the gradual change pattern into the second mapping function to obtain the risk score corresponding to the gradual change pattern. The second mapping function also adopts a piecewise non-linear design, but its slope change characteristics are different from those of the first mapping function to reflect the different ways in which the gradual change pattern and the periodic mutation pattern affect the risk.

[0050] Based on the change characteristics of physiological indicators within the time window, dynamically adjust the weight coefficients of the risk scores of the periodic mutation and the gradual change pattern. When severe periodic fluctuations are observed, increase the weight of the risk score of the periodic mutation; when a continuous gradual change trend is observed, increase the weight of the risk score of the gradual change pattern. Sum the weighted risk scores of the two types to obtain the comprehensive risk score.

[0051] Perform time integration on the comprehensive risk score to obtain the risk accumulation curve. The integration process adopts an adaptive step size design, using a smaller integration step size in the interval where the risk score changes violently and a larger integration step size in the interval where the risk score is relatively stable. The obtained risk accumulation curve can accurately reflect the accumulation process of heart failure risk over time. For example, for the monitoring data of a patient with congestive heart failure, if frequent heart rate mutations and continuous blood pressure drops are observed at night, the risk accumulation rate during this period will be significantly higher than other periods.

[0052] Through the above technical solutions, it is possible to comprehensively analyze the mutation characteristics and gradual change characteristics of physiological indicators, realize the accurate quantification and dynamic tracking of heart failure risk. This solution is particularly suitable for heart failure patients who need long-term continuous monitoring, can timely detect abnormal patterns during the risk accumulation process, and provide a decision-making basis for clinical intervention.

[0053] Traditional techniques usually detect mutation points through simple threshold judgment or static feature extraction. However, these methods are prone to ignoring complex change patterns in the data, resulting in insufficient sensitivity to mutations. In this application, an adaptive weight function is introduced to construct a mutation topology network. By combining time interval and amplitude information, the weight assignment to mutation points can be dynamically adjusted, enabling a more accurate quantification of the correlation between mutation points. In addition, the application of local entropy and enhanced topological entropy sequences can capture the detailed changes in mutation data at multiple scales, effectively identifying periodic and gradual mutation patterns. In the prior art, usually only fixed-scale or single-frequency domain analysis is relied on, making it difficult to comprehensively reveal the diversity and complexity of the data. In this application, through optimizing scale selection and dynamic window adjustment, periodic and gradual features can be continuously and stably identified in a changing environment, improving the accuracy of pattern recognition. The calculation of the risk score transforms the duration and frequency of periodic and gradual patterns through a multi-level mapping function. Combining dynamic weighting, it can comprehensively reflect the severity and potential risks of mutations. Compared with traditional risk assessment methods, this solution can provide a more comprehensive and accurate risk assessment, avoiding overly simple processing methods, thereby improving the reliability and adaptability of mutation point detection and risk assessment.

[0054] Figure 3 It is a performance comparison diagram of the risk assessment and early warning system according to an embodiment of the present invention. As Figure 3 shown, this diagram shows the performance comparison of the risk assessment and early warning system. The upper half of the chart is a comparison of five key indicators: in terms of accuracy, this technical solution reaches 95.4%, far higher than the threshold detection method (65.0%) and the statistical clustering method (72.5%); in terms of recall rate, this technical solution is 93.2%, while the threshold detection method and the statistical clustering method are 60.8% and 68.5% respectively; in terms of F1 score, the advantage of this technical solution is more obvious, reaching 96.2%, leading the threshold detection method (62.8%) and the statistical clustering method (70.4%) by more than 25 percentage points; in terms of the key indicator of early warning time, this technical solution can issue an early warning 186.5 seconds in advance, which is 3.2 times that of the threshold detection method (58.2 seconds) and 2 times that of the statistical clustering method (92.4 seconds); in terms of false alarm rate, this technical solution is only 3.8%, while the threshold detection method and the statistical clustering method are as high as 28.5% and 19.2% respectively.

[0055] In an alternative embodiment, the local entropy of nodes is calculated at different scales of the mutation topology network, and the optimal scale set is determined according to the change characteristics of the local entropy. The fusion of the local entropy under the optimal scale set to obtain the enhanced topological entropy sequence includes: Construct a node neighborhood matrix in the mutant topological network. The node neighborhood matrix represents the set of nodes reachable by each node within a specified number of steps. Based on the node neighborhood matrix, construct a node feature vector including degree centrality and clustering coefficient, calculate the node structure similarity, and adaptively reconstruct the network connection using the node structure similarity; Calculate the cumulative weight between nodes in the reconstructed network. The cumulative weight is obtained by cumulative calculation at different steps in a recursive manner. Based on the cumulative weight, calculate the transition probability between nodes, and adaptively and non-linearly adjust the transition probability according to the local density distribution of the network to obtain the corrected transition probability; Calculate the local entropy of nodes at different scales based on the corrected transition probability to obtain a local entropy sequence, calculate the multi-order difference features and community evolution features of the local entropy sequence, and construct a scale importance index; Determine the optimal scale set according to the scale importance index, calculate the information redundancy of each scale in the optimal scale set based on the complementary measure, take the normalized reciprocal of the information redundancy as the fusion weight, and perform weighted combination on the local entropy sequence under the optimal scale set using the fusion weight to obtain an enhanced topological entropy sequence.

[0056] The present invention provides a method for calculating the local entropy of nodes at different scales in a mutant topological network. First, construct a node neighborhood matrix in the mutant topological network. Taking a social network with 10 nodes as an example, construct the node neighborhood matrix by recording the set of nodes reachable by each node within different steps. For example, for node 1, the set of nodes reachable within 1 step is {2, 3, 5}, and the set of nodes reachable within 2 steps is {2, 3, 4, 5, 7}, and so on to construct the complete node neighborhood matrix.

[0057] Construct a node feature vector based on the node neighborhood matrix, including degree centrality and clustering coefficient. The degree centrality is obtained by calculating the number of direct connections of the node. For example, the degree centrality of node 1 is 3; the clustering coefficient is obtained by calculating the connection density between the neighbors of the node. For example, the clustering coefficient of node 1 is 0.67, indicating the ratio of the actual number of connections between its neighbor nodes to the total possible number of connections. Calculate the node structure similarity, and use the cosine similarity to measure the structural similarity between nodes. For nodes i and j, extract their feature vectors and calculate the cosine value of the vector angle as the similarity. For example, the structure similarity between node 1 and node 2 is 0.85, indicating that they have similar network positions and connection patterns.

[0058] Adaptive reconstruction of network connections is carried out using node structure similarity. When the structure similarity between two nodes exceeds the threshold of 0.8, an additional weight of 0.2 is added to the original connection weight; when the similarity is lower than 0.3, the original connection weight is reduced by 10%. In this way, the connection strength between nodes with similar structures is enhanced, and the connection between nodes with significant differences is weakened, forming a network structure that more conforms to the actual interaction logic. Calculate the cumulative weight between nodes in the reconstructed network. First, initialize the cumulative weight matrix. For directly connected nodes, the cumulative weight is equal to the connection weight between them; for indirectly connected nodes, it is obtained by recursively accumulating at different steps. For example, the two-step cumulative weight from node 1 to node 4 is 0.35, indicating the cumulative connection strength through all possible two-step paths.

[0059] Calculate the transition probability between nodes based on the cumulative weight. For node i, calculate the transition probability from i to all other nodes j, that is, normalize the cumulative weight starting from i. For example, the transition probability distribution from node 1 to each node is: node 2 (0.25), node 3 (0.2), node 4 (0.15), etc.

[0060] Carry out adaptive non-linear adjustment of the transition probability according to the local density distribution of the network to obtain the corrected transition probability. First, calculate the local density of each node, that is, the density of connections within the node's neighborhood, and then design an adjustment function: for nodes in high-density regions, reduce their probability of transferring to low-density regions; for nodes in low-density regions, increase their probability of transferring to high-density regions. For example, node 1 is located in a high-density region (density value 0.8), and its transition probability to node 6 in the low-density region is adjusted from the original 0.1 to 0.08.

[0061] Calculate the local entropy of nodes at different scales based on the corrected transition probability. The scale parameter controls the range of random walk to explore the network, and 10 different scales are set from 1 to 10. At each scale, calculate the local entropy of the node, which represents the complexity of the structure around the node. For example, the local entropy of node 1 at scale 2 is 1.58, and at scale 5 is 2.36.

[0062] Calculate the multi-order difference characteristics and community evolution characteristics of the local entropy sequence, and construct a scale importance index. The multi-order difference characteristics are obtained by calculating the change rate of the local entropy at different scales, and the community evolution characteristics are obtained by monitoring the stability of the community structure at different scales. For example, when the scale changes from 3 to 4, the average change rate of the local entropy of the node is 15%, and the stability of the community structure decreases by 12%. The comprehensive importance index for this scale conversion is 0.78.

[0063] Determine the optimal scale set according to the scale importance index. Select the scales corresponding to the scale conversion points with the top 30% importance index rankings. For example, the optimal scale set selected in this example is {2, 4, 7, 9}.

[0064] Calculate the information redundancy of each scale in the optimal scale set based on the complementarity measure. Information redundancy measures the degree of overlap between the local entropy information at one scale and the information at other scales. For example, the average mutual information between scale 2 and other optimal scales is 0.4, so its information redundancy is 0.4; similarly, the information redundancy of scale 4 is 0.35; scale 7 is 0.3; scale 9 is 0.25. Take the normalized reciprocal of the information redundancy as the fusion weight. The calculation process is as follows: First, take the reciprocal to get {2.5, 2.86, 3.33, 4}, and then normalize to get the final fusion weights {0.2, 0.225, 0.26, 0.315}.

[0065] Use the fusion weights to perform weighted combination on the local entropy sequences under the optimal scale set to obtain the enhanced topological entropy sequence. For example, for node 1, its local entropy values under the scale set {2, 4, 7, 9} are {1.58, 2.36, 2.85, 3.12} respectively. After weighted combination, the enhanced topological entropy value is 2.58. This enhanced topological entropy sequence synthesizes the structural information at multiple scales and can more accurately characterize the importance and influence of nodes in the network.

[0066] Based on the above technical solutions, it is possible to achieve refined analysis of the mutant topological network and enhanced feature extraction. Traditional network analysis methods mostly rely on single-scale or fixed node structure features for analysis, which easily ignores the multi-dimensional and complex interrelationships in the data, resulting in inaccurate capture of mutant features in the network. This application can more comprehensively reflect the structural relationship between nodes by constructing a node neighborhood matrix and a node feature vector, combining degree centrality and clustering coefficient, and then adaptively reconstruct and optimize the network connection through adaptive reconstruction, making the overall structure of the network more flexible and adaptable.

[0067] On this basis, the calculation of recursive cumulative node weights and modified transition probabilities can more accurately represent the relationship strength and information flow between nodes, making up for the deficiencies of traditional methods that ignore dynamic changes and non-linear regulation. In addition, the introduction of multi-order difference features and community evolution features of the local entropy sequence enables the identification of mutation points not to be limited to single-level analysis, but to comprehensively consider the evolution and information changes of nodes from multiple dimensions and levels.

[0068] Compared with the prior art, by dynamically selecting the optimal scale set and introducing the weight adjustment of information redundancy, the fused local entropy sequence can more accurately reflect the core features of mutations, while avoiding the interference of redundant information. The enhanced topological entropy sequence obtained by weighted combination can comprehensively consider the diversity of mutation features at different scales, improve the adaptability to complex data structures, and enhance the ability to identify mutation patterns.

[0069] In an alternative embodiment, the change time sequence relationship between physiological indicators is calculated according to the risk accumulation curve, the change time sequence relationship is converted into the index conduction delay, the risk conduction path diagram is drawn based on the index conduction delay, and the time interval with the largest change slope is marked in the risk conduction path diagram as the intervention time window, including: Extract the local features of the risk accumulation curve at multiple time window scales, calculate the distance matrix between physiological indicators according to the local features, and calculate the change time sequence relationship between the physiological indicators by using the time warping algorithm with attention weights based on the distance matrix; Construct the hierarchical coupling network between the physiological indicators, recursively calculate the direct coupling coefficient and the indirect coupling coefficient in the hierarchical coupling network, and perform weighted correction on the change time sequence relationship to obtain the index conduction delay; Construct a risk conduction path diagram with the physiological indicators as nodes and the index conduction delay as edge weights, calculate the dynamic scores of the nodes in the risk conduction path diagram by using the time-varying node importance algorithm, and determine the conduction path according to the dynamic scores and the index conduction delay; Adopt an adaptive segmented planning algorithm on the conduction path to search for the interval with the largest slope to obtain a candidate time window, calculate the information entropy of the slope sequence within the candidate time window to obtain the window confidence level. When the window confidence level is lower than the preset confidence threshold, expand the interval range along the conduction path to re-search for the interval with the largest slope, and use the interval with the largest slope that satisfies the window confidence level as the final intervention time window.

[0070] Exemplarily, extract the local features of the risk accumulation curve at multiple time window scales. Specifically, 5 minutes, 15 minutes, 30 minutes, 60 minutes, and 120 minutes can be selected as different time window scales. For the risk accumulation curve of each physiological indicator, calculate the statistical features at each time window scale respectively, including mean, variance, kurtosis, skewness, maximum value, minimum value, and volatility. For example, for the blood pressure indicator, the mean of 120 mmHg, variance of 5.2, kurtosis of 2.1, skewness of 0.8, maximum value of 135 mmHg, minimum value of 105 mmHg, and volatility of 0.12 can be obtained in the 5-minute window; the mean of 118 mmHg, variance of 4.8, etc. corresponding eigenvalue can be obtained in the 15-minute window. Similarly, the same feature extraction process is performed on other physiological indicators such as blood sugar and heart rate.

[0071] Calculate the distance matrix between physiological indicators based on local features. Use the Euclidean distance to calculate the distance between the eigenvectors of each indicator at different time window scales, and form the original distance matrix. For example, the distance between blood pressure and heart rate indicators at the 5-minute window is 2.35, and the distance at the 15-minute window is 2.89. Then, introduce an attention weight mechanism to assign different weights to the distances at different time window scales. Specifically, by calculating the variance of the distances at each window scale, the window with a larger variance is given a higher weight. For example, the weight of the 5-minute window is 0.15, the weight of the 15-minute window is 0.20, the weight of the 30-minute window is 0.25, the weight of the 60-minute window is 0.30, and the weight of the 120-minute window is 0.10. Finally, obtain the comprehensive distance matrix through weighted average.

[0072] Based on the distance matrix, use the time warping algorithm with attention weights to calculate the change time series relationship between physiological indicators. This algorithm first constructs a time alignment path, then identifies key alignment points through the attention mechanism, and finally obtains the change time series relationship between indicators. For example, the calculation results show that the blood pressure change precedes the heart rate change by 8.5 minutes, the heart rate change precedes the blood oxygen change by 12.3 minutes, and the blood oxygen change precedes the body temperature change by 5.2 minutes. Next, construct a hierarchical coupling network between physiological indicators. This network consists of three layers: the internal fluctuation layer of a single indicator, the direct coupling layer between indicators, and the indirect coupling layer between indicators. During the construction process, first establish the network edge connections based on the change correlation between indicators, and then assign edge weights. For example, the direct coupling coefficient between blood pressure and heart rate is 0.78, and the direct coupling coefficient between heart rate and blood oxygen is 0.65.

[0073] Recursively calculate the direct coupling coefficient and the indirect coupling coefficient in the hierarchical coupling network. The direct coupling coefficient is calculated by the Pearson correlation coefficient. For example, the direct coupling coefficient between blood pressure and heart rate is 0.78. The indirect coupling coefficient is calculated by path multiplication. For example, the coupling coefficient of blood pressure indirectly affecting blood oxygen through heart rate is 0.78×0.65 = 0.507. Then, perform weighted correction on the change time series relationship to obtain the indicator conduction delay. After correction, the conduction delay from blood pressure change to heart rate change is 7.8 minutes, and the conduction delay from heart rate change to blood oxygen change is 11.5 minutes.

[0074] Construct a risk conduction path diagram with physiological indicators as nodes and indicator conduction time delays as edge weights. In this diagram, the direction of the edge represents the risk conduction direction, and the edge weight represents the conduction time delay. For example, the edge weight from the blood pressure node to the heart rate node is 7.8 minutes, indicating that it takes 7.8 minutes for the risk to conduct from blood pressure to heart rate. Use a time-varying node importance algorithm to calculate the dynamic scores of the nodes in the risk conduction path diagram. This algorithm comprehensively considers the node connectivity, betweenness centrality, and closeness centrality, and introduces a time decay factor to calculate the importance scores of the nodes at different time points. For example, 30 minutes before the onset of the disease, the score of the blood pressure node is 0.85, and the score of the heart rate node is 0.72; 15 minutes before the onset of the disease, the score of the blood pressure node is 0.79, and the score of the heart rate node is 0.83, indicating that as time goes by, the importance of the heart rate node exceeds that of the blood pressure node.

[0075] Determine the conduction path according to the dynamic scores and indicator conduction time delays. Select the path with the highest total score as the main risk conduction path, such as: blood pressure → heart rate → blood oxygen → body temperature, with a total score of 3.19.

[0076] On the determined conduction path, use an adaptive segmented planning algorithm to search for the interval with the largest slope to obtain the candidate time window. This algorithm first divides the risk accumulation curve on the conduction path into several segments, calculates the slope for each segment, and uses the dynamic programming method to find the continuous interval with the largest slope. For example, on the blood pressure → heart rate conduction path, it is found that the slope in the interval from 25 to 20 minutes before the onset of the disease is the largest, which is 0.045 / minute.

[0077] Calculate the information entropy of the slope sequence within the candidate time window to obtain the window confidence level. The lower the information entropy, the more stable the slope change and the higher the window confidence level. For example, the information entropy of the slope sequence in the interval from 25 to 20 minutes before the onset of the disease is 0.28, and the corresponding window confidence level is 0.72. Set the window confidence level threshold to 0.7. When the window confidence level is lower than the threshold, expand the interval range along the conduction path to re-search for the interval with the largest slope. For example, if the confidence level of a certain candidate window is 0.65, which is lower than the threshold of 0.7, then expand the search range from the original 5 minutes to 7 minutes, and recalculate to obtain that the interval with the largest slope is from 27 to 20 minutes before the onset of the disease, and the corresponding confidence level is 0.75, meeting the threshold requirement. Take the interval with the largest slope that meets the window confidence level as the intervention time window. In this embodiment, the determined intervention time window is from 27 to 20 minutes before the onset of the disease, indicating that medical intervention within this time window may be the most effective.

[0078] Based on the above technical solution, it is possible to achieve dynamic monitoring of physiological index changes and accurate identification of precise intervention timing. Traditional technologies usually adopt static methods to analyze the relationships between physiological indexes, which easily overlook the temporal sequence and dynamic conduction effects of changes between indexes, resulting in inaccurate judgment of intervention timing. By extracting the local features of the risk accumulation curve and combining with the time warping algorithm with attention weights, this application can more accurately capture the temporal sequence relationship of changes between physiological indexes, thereby revealing the time delay and mutual influence of index changes.

[0079] Furthermore, the constructed hierarchical coupling network enhances the modeling ability of complex interactions between indexes by recursively calculating direct and indirect coupling coefficients, making the calculation of index conduction time delay more comprehensive and accurate. This method can better reflect the multi-level coupling relationship between physiological indexes compared with traditional simple distance calculation. By constructing a risk conduction path diagram and applying a time-varying node importance algorithm, the scores of each node can be dynamically calculated to help identify the main path of risk conduction, avoiding the deficiencies of relying solely on fixed models or static structures in the past.

[0080] In terms of the identification of the intervention time window, this solution introduces an adaptive segmented planning algorithm and information entropy calculation, which can identify the critical moments in risk conduction according to the change of slope. The introduction of this process optimizes the limitations of using simple thresholds or single scales in traditional methods, and can automatically adjust the selection of the intervention time window according to dynamic changes to ensure that the intervention timing is more scientific and effective. Through the above improvements, the solution is superior to the existing technologies in terms of accuracy, sensitivity, and real-time performance, and can better support the early warning and intervention of physiological abnormalities.

[0081] In an optional implementation manner, constructing the hierarchical coupling network between the physiological indexes, recursively calculating the direct coupling coefficient and the indirect coupling coefficient in the hierarchical coupling network, and performing weighted correction on the change time sequence relationship to obtain the index conduction time delay includes: Performing sliding window segmentation on the change time sequence relationship between physiological indexes to obtain local time sequence segments, calculating the matching degree of the local time sequence segments by using the dynamic time warping algorithm to obtain a time sequence similarity matrix, performing multi-scale wavelet transform on the local time sequence segments for frequency domain decomposition, calculating the phase synchronization of different frequency components to obtain a frequency coupling degree matrix, and adaptively weighted fusing the time sequence similarity matrix and the frequency coupling degree matrix to obtain the coupling strength between indexes; Constructing an initial hierarchical coupling network based on the coupling strength between indexes, calculating the conditional mutual information between network nodes, performing causal entropy significance test based on the conditional mutual information to screen stable coupling edges, updating the initial hierarchical coupling network with the stable coupling edges to obtain an optimized hierarchical coupling network, and extracting the coupling strength of the network adjacent edges in the optimized hierarchical coupling network to obtain the direct coupling coefficient; Perform a depth - first search on the optimized hierarchical coupling network to identify candidate conduction links between physiological indicators, calculate the information entropy of the candidate conduction links, set an effective threshold based on the information entropy to screen out effective conduction links, perform recursive calculations on the effective conduction links, and use the ratio of the cumulative effect of the coupling strengths of each edge on the link to the path length as the indirect coupling coefficient; Calculate the comprehensive coupling coefficient based on the direct coupling coefficient and the indirect coupling coefficient, use the comprehensive coupling coefficient to perform non - linear weighted correction on the changing timing relationship, normalize the corrected changing timing relationship and convert it into time units to obtain the index conduction delay.

[0082] Exemplarily, in order to analyze the conduction relationship between physiological indicators, it is first necessary to perform a refined analysis of the changing timing relationship of each indicator. Segment the timing data by setting a sliding window, and the window size can be dynamically adjusted according to the signal characteristics. For example, for the heart rate variability index, a basic window size of five minutes can be set, and for the blood pressure index, a basic window size of ten minutes can be set. The sliding step is set to one - quarter of the window size to ensure sufficient overlap between adjacent windows to capture continuous change characteristics.

[0083] For each local timing segment within the sliding window, use the dynamic time warping algorithm to calculate the matching degree between different indicators. This algorithm can handle timing data of different lengths and different sampling rates by constructing a distance matrix and finding the optimal time alignment path. For example, when analyzing the correlation between heart rate and blood pressure, even if the sampling frequencies of the two signals are different, their best matching relationship can be obtained through dynamic warping.

[0084] Convert the result of dynamic time warping into a timing similarity matrix, where each element in the matrix represents the similarity degree of the corresponding indicator pair in the current time window. The similarity value is mapped to the range from zero to one through normalization, and the larger the value, the closer the change patterns of the two indicators. For example, if the heart rate and respiratory rate show highly synchronous changes within a certain time window, their similarity value will be close to one. At the same time, perform multi - scale wavelet transform on the local timing segment, and decompose the time - domain signal into different frequency components. The wavelet basis function is selected as the Ricker wavelet, and the decomposition scale ranges from one to six, covering the complete frequency spectrum range from high - frequency to low - frequency. In this way, the signal characteristics at different time scales can be analyzed. For example, the rapid fluctuations of the heart rate and the slow changes of the blood pressure can be captured simultaneously.

[0085] Calculate the phase synchrony between different frequency components to obtain the frequency coupling degree matrix. The phase synchrony extracts the instantaneous phase of the signal through the Hilbert transform and then calculates the stability of the phase difference. For example, when a stable phase relationship is observed between heart rate and blood oxygen saturation in a specific frequency band, it indicates that these two indicators have significant coupling characteristics at this frequency. Perform adaptive weighted fusion on the time series similarity matrix and the frequency coupling degree matrix to obtain the comprehensive coupling strength between indicators. The weighting coefficient is dynamically adjusted according to the signal-to-noise ratio of each indicator, and a larger weight is assigned to the indicator with better signal quality. The coupling strength obtained in this way contains both time-domain characteristics and frequency-domain characteristics and can comprehensively reflect the correlation between indicators.

[0086] Construct an initial hierarchical coupling network based on the coupling strength. The nodes in the network represent different physiological indicators, and the weight of the edge is determined by the coupling strength. To ensure the reliability of the network structure, it is necessary to calculate the conditional mutual information between network nodes. The conditional mutual information reflects the true dependence relationship between two nodes considering the influence of other nodes. Perform a causal entropy significance test on the conditional mutual information to screen out stable coupling edges. Specifically, by constructing a null hypothesis distribution of random permutations, calculate the significance level at which the actual observed value exceeds this distribution. For example, if the conditional mutual information between heart rate and blood pressure is significantly higher than the random level, then retain the coupling edge between these two indicators.

[0087] Update the initial network using the screened stable coupling edges to obtain an optimized hierarchical coupling network. Extract the coupling strength of the network adjacent edges in the optimized network as the direct coupling coefficient, which reflects the direct influence relationship between indicators. For example, a high direct coupling coefficient between heart rate and cardiac output indicates that there is a direct physiological regulatory relationship between these two indicators. Perform a depth-first search on the optimized hierarchical coupling network to identify possible conduction links between indicators. Record all possible paths during the search process to form a set of candidate conduction links. For example, there may be multiple conduction paths from heart rate to blood pressure, including direct paths and indirect paths through other indicators. Calculate the information entropy of the candidate conduction links to evaluate the information transfer efficiency of the links. The lower the information entropy, the more stable the conduction process and the higher the reliability of the link. Set a screening threshold based on the information entropy and retain the links with information entropy lower than the threshold as effective conduction links.

[0088] Perform recursive calculations on the effective conduction links, considering the cumulative effect of each coupling edge on the link. Take the ratio of the cumulative coupling strength to the link length as the indirect coupling coefficient, which reflects the efficiency of remote conduction. For example, if there is a stable conduction link from heart rate to tissue perfusion index passing through multiple intermediate nodes and its indirect coupling coefficient is still high, it indicates that this conduction path has important physiological significance.

[0089] Construct a comprehensive evaluation model based on the direct coupling coefficient and the indirect coupling coefficient, and calculate the comprehensive coupling coefficient. The model adopts a non-linear weighting method, with different weight allocation strategies in different coupling strength ranges. For example, when both the direct coupling and the indirect coupling are strong, the comprehensive coefficient will be significantly improved, reflecting the synergistic effect of multiple conduction pathways. Use the comprehensive coupling coefficient to non-linearly weight and correct the original change time series relationship. The correction process takes into account the complexity and stability of the conduction link, and can more accurately reflect the actual conduction delay between indicators. For example, if a certain conduction link has a high comprehensive coupling coefficient, the corresponding time delay estimate value will be reduced.

[0090] Normalize the corrected time series relationship to eliminate the influence of different indicator dimensions. Finally, convert the normalized time series relationship into actual time units to obtain the indicator conduction time delay. This time delay value intuitively reflects the time required for a change in one indicator to cause a response in another indicator, providing an important time window reference for clinical intervention.

[0091] Based on the above technical solutions, it is possible to achieve an in-depth understanding and accurate modeling of the complex coupling relationships between physiological indicators. By combining dynamic time warping and wavelet transform, it is not only possible to analyze the change time series between indicators from the time domain perspective, but also to explore the coupling degree of different frequency components in the frequency domain, which makes the calculation of the coupling strength more comprehensive and accurate. Further optimize the hierarchical coupling network through conditional mutual information and causal entropy significance tests, and screen out stable coupling edges, effectively improving the stability and reliability of the coupling relationship. In terms of conduction path identification, by using depth-first search and information entropy to screen effective links, it is possible to dynamically capture the key conduction paths between physiological indicators, avoiding complex path relationships that may be ignored in traditional methods. Recursively calculate the ratio of the cumulative effect to the path length, providing a new method for the accurate calculation of the indirect coupling coefficient, making the measurement of the conduction time delay more detailed and accurate. Through non-linear weighting correction and normalization processing, it is possible to accurately calculate the conduction time delay of physiological indicators, providing reliable data support for further intervention decisions. Overall, this solution improves the accuracy and adaptability of physiological data analysis, especially in the modeling of multi-scale, multi-frequency and non-linear relationships, showing obvious improvement effects compared with traditional methods.

[0092] Figure 4 This is a comparison chart of the time delay calculation accuracy for the embodiments of the present invention, as Figure 4As shown, this figure presents a comparison of the mean absolute errors in calculating time delays between six pairs of physiological indicators using four different methods. The proposed technical solution adopts a multi-dimensional coupling analysis and hierarchical coupling network method, demonstrating the lowest error values in all test cases. Specifically, in the blood pressure-heart rate indicator pair, the error of the proposed technical solution is only 0.22 seconds, while the errors of the cross-correlation method, phase synchronization method, and traditional coupling network method are 0.60 seconds, 0.80 seconds, and 0.40 seconds respectively; in the most complex indicator pair of electroencephalogram-blood pressure, the error of the proposed technical solution is 0.44 seconds, while the errors of the other three methods are 1.10 seconds, 1.40 seconds, and 0.80 seconds respectively, showing a more significant advantage. Notably, the traditional cross-correlation method (based on signal correlation analysis) and phase synchronization method (based on phase extraction using Hilbert transform) perform poorly in dealing with non-linear coupling relationships, while the traditional coupling network method (such as Granger causal network), although improved, does not consider frequency domain characteristics, thus having limited accuracy. The proposed technical solution significantly improves the accuracy of calculating time delays between various physiological indicators by integrating time domain and frequency domain features and performing information entropy screening.

[0093] In the second aspect of the embodiments of the present invention, there is provided a schematic structural diagram of a heart failure risk assessment system based on AI, where the system includes: A first unit configured to collect physiological indicators of an object to be evaluated to construct a multi-dimensional time series data stream, calculate the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm, determine the time points with a difference degree greater than a preset difference threshold as mutation points, extract the waveform features of the data segments before and after the mutation points, and generate a mutation feature matrix; A second unit configured to perform a time series mapping on the mutation feature matrix, obtain the time interval sequence and amplitude sequence of the mutation features, identify periodic mutation patterns and gradual change patterns, convert the duration and occurrence frequency of the periodic mutation patterns and gradual change patterns into risk scores, and draw a risk accumulation curve; A third unit configured to calculate the change time series relationship between physiological indicators according to the risk accumulation curve, convert the change time series relationship into an index conduction time delay, draw a risk conduction path diagram based on the index conduction time delay, and mark the time interval with the largest change slope in the risk conduction path diagram as an intervention time window; A fourth unit configured to determine the heart failure risk level based on the time range of the intervention time window and the change trend of the risk accumulation curve, and generate a risk assessment report including the risk level and abnormal description of physiological indicators.

[0094] In the third aspect of the embodiments of the present invention, there is provided an electronic device, including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0095] In a fourth aspect of the embodiments of the present invention, there is provided a computer-readable storage medium, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0096] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium, on which computer-readable program instructions for executing various aspects of the present invention are loaded.

[0097] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI-based heart failure risk assessment method, characterized in that, Including: Collect the physiological indicators of the object to be evaluated and construct them into a multi-dimensional time series data stream. Calculate the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm. Determine the time points with a difference degree greater than the preset difference threshold as mutation points, extract the waveform features of the data segments before and after the mutation points, and generate a mutation feature matrix. Perform a time series mapping on the mutation feature matrix, obtain the time interval sequence and amplitude sequence of the mutation features, identify the periodic mutation pattern and the gradual change pattern, convert the duration and occurrence frequency of the periodic mutation pattern and the gradual change pattern into risk scores, and draw a risk accumulation curve. Calculate the changing time series relationship between physiological indicators according to the risk accumulation curve, convert the changing time series relationship into the index conduction delay, draw a risk conduction path diagram based on the index conduction delay, and mark the time interval with the largest change slope in the risk conduction path diagram as the intervention time window. Based on the time range of the intervention time window and the changing trend of the risk accumulation curve, determine the heart failure risk level, and generate a risk assessment report including the risk level and the description of abnormal physiological indicators.

2. The method according to claim 1, characterized in that Collect the physiological indicators of the object to be evaluated and construct them into a multi-dimensional time series data stream. Calculating the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm and determining the time points with a difference degree greater than the preset difference threshold as mutation points includes: Collect the physiological indicators of the object to be evaluated, where the physiological indicators include cardiovascular data, respiratory data, and blood oxygen data; resample the physiological indicators through cubic spline interpolation to construct a multi-dimensional time series data stream, and construct the multi-dimensional time series data stream into a state matrix in the time-index dimension. Make an initial division of the state matrix from the midpoint position to obtain two sub-data segments. Move the division position point by point along the time axis to calculate the Mahalanobis distance between the two data segments on both sides to obtain a distance sequence and perform eigenvalue decomposition to obtain the principal component weight vector. Use the principal component weight vector to weight the distance sequence to obtain the sub-data segment difference degree sequence. Statistically analyze the probability distribution characteristics of the sub-data segment difference degree sequence to obtain a cumulative distribution function. Determine the difference threshold on the cumulative distribution function according to the preset false alarm rate. Mark the time points greater than the difference threshold as initial mutation points. Construct an observation window centered on the initial mutation points, calculate the mutual information between the signals of each dimension within the observation window to obtain a mutual information matrix, and use the sum of the non-diagonal elements of the mutual information matrix as the coupling metric value. Calculate the comprehensive difference degree based on the coupling metric value and the sub-data segment difference degree sequence, screen the initial mutation points according to the comprehensive difference degree, and determine the time points with a comprehensive difference degree greater than the difference threshold as the final mutation points.

3. The method according to claim 1, wherein Extracting the waveform features of the data segments before and after the mutation points and generating a mutation feature matrix includes: Calculate the fluctuation characteristics of the data segments before and after the mutation points according to the mutation point position to obtain a volatility sequence, and perform segmented cumulative calculation to obtain the fluctuation trend feature. Determine the adaptive window length based on the fluctuation trend feature, and use the adaptive window length to determine the waveform feature extraction interval before and after the mutation points. Within the waveform feature extraction interval, perform time-frequency transformation on the data segments before and after the mutation point respectively to obtain time-frequency spectra, calculate the energy distribution of the time-frequency spectra to obtain a time-frequency energy sequence, perform weighted fusion on the time-frequency energy sequence in combination with the volatility sequence to obtain a fusion feature sequence, and calculate the multi-scale entropy value of the fusion feature sequence to obtain a complexity feature; Combine the volatility sequence, time-frequency energy sequence, and complexity feature in chronological order to obtain a feature vector, construct an initial feature matrix based on the feature vector, calculate the discriminant contribution degree of each feature in the initial feature matrix to obtain feature weights, and use the feature weights to perform weighted reconstruction on the initial feature matrix to generate a mutation feature matrix.

4. The method according to claim 1, characterized in that, Perform chronological mapping on the mutation feature matrix, obtain the time interval sequence and amplitude sequence of the mutation features, identify periodic mutation patterns and gradual change patterns, convert the duration and occurrence frequency of the periodic mutation patterns and gradual change patterns into risk scores, and draw a risk accumulation curve including: Perform chronological mapping on the mutation feature matrix to obtain a time interval sequence and an amplitude sequence, and construct a mutation topology network through an adaptive weight function based on the time interval sequence and the amplitude sequence, where nodes represent mutation points, and the connection weights between nodes are jointly determined by the time interval and the amplitude; Calculate the local entropy of nodes at different scales of the mutation topology network, determine the optimal scale set according to the change characteristics of the local entropy, and fuse the local entropy under the optimal scale set to obtain an enhanced topological entropy sequence; Dynamically adjust the calculation window size based on the change rate of the enhanced topological entropy sequence, calculate the autocorrelation function and power spectral density of the enhanced topological entropy sequence within the calculation window, extract periodic mutation features, determine the first duration and the first occurrence frequency of the periodic mutation pattern according to the periodic mutation features, perform piecewise polynomial fitting on the enhanced topological entropy sequence to identify the gradual change interval, and determine the second duration and the second occurrence frequency of the gradual change pattern according to the gradual change interval; Convert the first duration and the first occurrence frequency into a first risk score through a first mapping function, convert the second duration and the second occurrence frequency into a second risk score through a second mapping function, obtain a risk score according to the dynamic weighted combination of the first risk score and the second risk score, and perform time integration on the risk score to obtain a risk accumulation curve.

5. The method according to claim 4, wherein Calculate the local entropy of nodes at different scales of the mutation topology network, determine the optimal scale set according to the change characteristics of the local entropy, and fuse the local entropy under the optimal scale set to obtain an enhanced topological entropy sequence including: Construct a node neighborhood matrix in the mutation topology network, where the node neighborhood matrix represents the set of nodes reachable by each node within a specified number of steps, construct a node feature vector including degree centrality and clustering coefficient based on the node neighborhood matrix, calculate the node structure similarity, and perform adaptive reconstruction on the network connection using the node structure similarity; Calculate the cumulative weight between nodes in the reconstructed network. The cumulative weight is obtained by cumulative calculation at different step numbers in a recursive manner. Calculate the transition probability between nodes based on the cumulative weight, and perform adaptive non-linear adjustment on the transition probability according to the local density distribution of the network to obtain the corrected transition probability; Calculate the local entropy of nodes at different scales based on the corrected transition probability to obtain a local entropy sequence. Calculate the multi-order difference features and community evolution features of the local entropy sequence, and construct a scale importance index; Determine the optimal scale set according to the scale importance index. Calculate the information redundancy of each scale in the optimal scale set based on the complementary measure. Take the normalized reciprocal of the information redundancy as the fusion weight, and use the fusion weight to perform weighted combination on the local entropy sequence under the optimal scale set to obtain an enhanced topological entropy sequence.

6. The method according to claim 1, wherein Calculate the change time series relationship between physiological indicators according to the risk accumulation curve, convert the change time series relationship into the index conduction delay, draw a risk conduction path diagram based on the index conduction delay, and mark the time interval with the largest change slope in the risk conduction path diagram as the intervention time window, including: Extract the local features of the risk accumulation curve at multiple time window scales, calculate the distance matrix between physiological indicators according to the local features, and calculate the change time series relationship between the physiological indicators by using a time warping algorithm with attention weights based on the distance matrix; Construct a hierarchical coupling network between the physiological indicators. Recursively calculate the direct coupling coefficient and the indirect coupling coefficient in the hierarchical coupling network, and perform weighted correction on the change time series relationship to obtain the index conduction delay; Construct a risk conduction path diagram with the physiological indicators as nodes and the index conduction delay as edge weights. Calculate the dynamic score of the nodes in the risk conduction path diagram by using a time-varying node importance algorithm, and determine the conduction path according to the dynamic score and the index conduction delay; Adopt an adaptive segmented planning algorithm on the conduction path to search for the interval with the largest slope to obtain a candidate time window. Calculate the information entropy of the slope sequence within the candidate time window to obtain the window confidence level. When the window confidence level is lower than the preset confidence threshold, expand the interval range along the conduction path to re-search for the interval with the largest slope, and take the interval with the largest slope that meets the window confidence level as the final intervention time window.

7. The method according to claim 6, wherein Construct a hierarchical coupling network between the physiological indicators. Recursively calculate the direct coupling coefficient and the indirect coupling coefficient in the hierarchical coupling network, and perform weighted correction on the change time series relationship to obtain the index conduction delay, including: Perform sliding window segmentation on the change time series relationship between physiological indicators to obtain local time series segments. Calculate the matching degree of the local time series segments by using a dynamic time warping algorithm to obtain a time series similarity matrix. Perform multi-scale wavelet transform on the local time series segments for frequency domain decomposition, calculate the phase synchrony of different frequency components to obtain a frequency coupling degree matrix, and perform adaptive weighted fusion on the time series similarity matrix and the frequency coupling degree matrix to obtain the coupling strength between indicators; Construct an initial hierarchical coupling network based on the coupling strength between indicators, calculate the conditional mutual information between network nodes, perform a causal entropy significance test based on the conditional mutual information to screen for stable coupling edges, use the stable coupling edges to update the initial hierarchical coupling network to obtain an optimized hierarchical coupling network, and extract the coupling strength of the network adjacent edges in the optimized hierarchical coupling network to obtain a direct coupling coefficient; Perform a depth-first search on the optimized hierarchical coupling network to identify candidate conduction links between physiological indicators, calculate the information entropy of the candidate conduction links, set an effective threshold based on the information entropy to screen for effective conduction links, perform recursive calculations on the effective conduction links, and use the ratio of the cumulative effect of the coupling strength of each edge on the link to the path length as an indirect coupling coefficient; Calculate a comprehensive coupling coefficient based on the direct coupling coefficient and the indirect coupling coefficient, use the comprehensive coupling coefficient to perform non-linear weighted correction on the changing time series relationship, normalize the corrected changing time series relationship and convert it into time units to obtain the indicator conduction delay.

8. An AI-based heart failure risk assessment system for implementing the method according to any one of the preceding claims 1-7, characterized in that, Includes: The first unit is used to collect the physiological indicators of the object to be evaluated and construct them into a multi-dimensional time series data stream. Calculate the difference degree between adjacent data segments in the multi-dimensional time series data stream through a recursive segmentation algorithm. Determine the time points with a difference degree greater than a preset difference threshold as mutation points, extract the waveform characteristics of the data segments before and after the mutation points, and generate a mutation feature matrix; The second unit is used to perform a time series mapping on the mutation feature matrix, obtain the time interval sequence and amplitude sequence of the mutation features, identify periodic mutation patterns and gradual change patterns, convert the duration and occurrence frequency of the periodic mutation patterns and gradual change patterns into risk scores, and draw a risk accumulation curve; The third unit is used to calculate the changing time series relationship between physiological indicators according to the risk accumulation curve, convert the changing time series relationship into an indicator conduction delay, draw a risk conduction path diagram based on the indicator conduction delay, and mark the time interval with the largest change slope in the risk conduction path diagram as the intervention time window; The fourth unit is used to determine the heart failure risk level based on the time range of the intervention time window and the changing trend of the risk accumulation curve, and generate a risk assessment report including the risk level and abnormal description of physiological indicators.

9. An electronic device, characterized in that, Includes: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

Images