Death risk assessment system for heart failure patients based on deep learning

Abstract

This invention relates to the field of medical informatics technology, specifically to a deep learning-based mortality risk assessment system for heart failure patients. The system includes a memory and a processor. The processor executes a computer program stored in the memory to perform the following steps: acquiring electrocardiogram (ECG) data and respiratory rate data of the target individual; determining the stationarity based on data differences and time intervals in the ECG data; filtering the mean of the ECG data using time windows of different scales to obtain data sequences at each scale; obtaining heart failure cycle-specific values ​​based on data changes around the maxima in the data sequences at each scale, the time interval between adjacent maxima, and the correlation between the data sequences and respiratory rate data; and determining error parameters by combining the sample entropy of the data sequences at each scale and the differences in sample entropy at different scales, thereby obtaining the level of attention given to the target individual. This invention improves the accuracy and reliability of the results in determining the level of attention given to the target individual.

Description

Technical Field

[0001] This invention relates to the field of medical informatics technology, specifically to a deep learning-based mortality risk assessment system for heart failure patients. Background Technology

[0002] Heart failure is a complex clinical syndrome caused by structural or functional abnormalities of the heart that prevent the cardiac output from meeting the body's metabolic needs, and it has become a significant public health burden. Accurate assessment of a patient's risk of death is crucial for clinical decision-making, directly impacting treatment options (such as drug optimization, device implantation, or heart transplantation), resource allocation, and the timing of palliative care intervention.

[0003] In existing technologies, risk assessment often relies on static clinical characteristics (such as age and medical history) or a single dynamic physiological indicator, lacking an integrated analysis of real-time dynamic changes in cardiac function and basic health status. This makes it difficult to comprehensively reflect the multidimensional risks of disease progression in patients. Furthermore, analysis of RR intervals based on a single time scale (such as focusing only on short-term heart rate variability) cannot fully characterize the multidimensional abnormalities of cardiac rhythm in heart failure patients (such as periodic oscillations at different time scales and rhythm disturbances caused by sympathetic hyperactivity). Consequently, it is impossible to accurately determine the level of attention of the person being tested, resulting in untimely subsequent risk assessment. Summary of the Invention

[0004] To address the problem that existing methods cannot accurately determine the level of attention of individuals being tested, the present invention aims to provide a deep learning-based mortality risk assessment system for heart failure patients. The specific technical solution adopted is as follows:

[0005] This invention provides a deep learning-based mortality risk assessment system for heart failure patients. The system includes a memory and a processor, the processor executing a computer program stored in the memory to achieve the following steps:

[0006] Obtain electrocardiogram and respiratory rate data of the target personnel;

[0007] Based on the differences and time intervals of data at different locations on the electrocardiogram (ECG) data, the stability is obtained; mean filtering is performed on each data point on the ECG data using time windows of different scales to obtain data sequences at each scale; based on the changes in data on both sides of the maximum point in the data sequence at each scale, the time interval between adjacent maximum points, the stability, and the correlation between the data sequence at each scale and the respiratory rate data, the heart failure cycle-specific values ​​of the data sequence at each scale are obtained.

[0008] Based on the sample entropy of the data sequence at each scale, the difference between sample entropy at different scales, and the heart failure cycle-specific value, the error parameters of the data sequence at each scale are determined.

[0009] The level of attention to the target personnel is determined using the error parameter and the sample entropy.

[0010] Preferably, obtaining the stability based on data differences and time intervals at different locations on the electrocardiogram data includes:

[0011] The difference between adjacent data points in the electrocardiogram data is recorded as the first difference.

[0012] The stability is obtained based on the difference between each first difference and the average of all first differences, the time interval between the maximum and minimum values ​​on the electrocardiogram data, and the numerical difference between the maximum and minimum values ​​on the electrocardiogram data.

[0013] Preferably, the step of obtaining the stability based on the difference between each first difference and the average of all first differences, the time interval between the maximum and minimum values ​​on the electrocardiogram data, and the numerical difference between the maximum and minimum values ​​on the electrocardiogram data includes:

[0014] Calculate a first ratio between the time interval and the numerical difference;

[0015] The stability is obtained based on the difference between each first difference and the average of all first differences and the first ratio. The difference between each first difference and the average of all first differences is negatively correlated with the stability, and the first ratio is positively correlated with the stability.

[0016] Preferably, the step of obtaining the heart failure cycle-specific values ​​of the data sequence at each scale based on the data changes on both sides of the maximum point in the data sequence at each scale, the time interval between adjacent maximum points, the degree of stability, and the correlation between the data sequence at each scale and the respiratory rate data includes:

[0017] For any scale:

[0018] Based on the difference in the degree of data change between the left and right sides of each maximum point at any given scale, the number of maximum points, and the time interval between adjacent maximum points, the oscillation performance value of the data sequence at any given scale is obtained.

[0019] The correlation between the data sequence at any scale and the respiratory rate data sequence is denoted as the first correlation corresponding to any scale. The respiratory rate data sequence is obtained by sorting the respiratory rate data in chronological order.

[0020] The respiratory correlation degree corresponding to any of the first correlation scales is obtained;

[0021] By combining the oscillatory performance value, the respiratory correlation, and the stability, the heart failure cycle-specific value of the data sequence at any scale is determined. The oscillatory performance value is positively correlated with the heart failure cycle-specific value, while the respiratory correlation and the stability are both negatively correlated with the heart failure cycle-specific value.

[0022] Preferably, obtaining the oscillation performance value of the data sequence at any scale based on the difference in the degree of data change on the left and right sides of each maximum point at any scale, the number of maximum points, and the time interval between adjacent maximum points includes:

[0023] Calculate the first average slope value of each data point on each side of each maximum point at any given scale;

[0024] The difference between the first average slope values ​​corresponding to both sides of each maximum point at any scale is recorded as the second difference of each maximum point at any scale.

[0025] Calculate a second ratio between the number of maxima in the data sequence at any given scale and the average time interval between all adjacent maxima.

[0026] Based on the second difference and the second ratio of all maxima at any given scale, the oscillation performance value of the data sequence at any given scale is obtained. The second difference is negatively correlated with the oscillation performance value, and the second ratio is positively correlated with the oscillation performance value.

[0027] Preferably, obtaining the respiratory correlation degree corresponding to any scale based on the first correlation includes:

[0028] The difference between the length of the time window at any scale and the length of the time window at the scale corresponding to the largest first correlation is denoted as the third difference;

[0029] The product of the third difference and the first correlation corresponding to any scale is determined as the respiratory correlation corresponding to any scale.

[0030] Preferably, determining the heart failure cycle-specific value of the data sequence at any scale by combining the oscillatory performance value, the respiratory correlation, and the stability includes:

[0031] Calculate the sum of the respiratory correlation coefficient and the zero-prevention parameter, where the zero-prevention parameter is greater than 0;

[0032] Calculate the third ratio of the oscillation performance value to the sum value, and determine the product of the reciprocal of the stability degree and the third ratio as the heart failure cycle-specific value of the data sequence at any scale.

[0033] Preferably, determining the error parameters of the data sequence at each scale based on the sample entropy of the data sequence at each scale, the difference between sample entropies at different scales, and the heart failure cycle-specific value includes:

[0034] Sort the sample entropy of the data sequences at all scales in ascending order to obtain the sample entropy sequence, and then perform curve fitting on the data in the sample entropy sequence to obtain the sample entropy curve.

[0035] For any scale: Based on the slope value of the sample entropy curve of the given scale, the sample entropy of the data sequence at the given scale, and the heart failure cycle-specific value, the error parameters of the data sequence at the given scale are obtained.

[0036] Preferably, obtaining the error parameters of the data sequence at any scale based on the slope value of the sample entropy curve at any scale, the sample entropy of the data sequence at any scale, and the heart failure cycle-specific value includes:

[0037] The ratio of the absolute value of the slope of the sample entropy curve at any scale to the sample entropy of the data sequence at any scale is denoted as the third ratio; the product of the third ratio and the heart failure cycle specificity of the data sequence at any scale is determined as the error parameter of the data sequence at any scale.

[0038] Preferably, determining the level of attention to the target personnel using the error parameter and the sample entropy includes:

[0039] Sort all scales in descending order of sample entropy to obtain a scale sequence; use the error parameters of the data sequences at the first preset number of scales in the scale sequence as key parameters;

[0040] The key parameters are input into the trained neural network to obtain predicted values;

[0041] The level of attention given to the target personnel is determined based on the predicted values.

[0042] The present invention has at least the following beneficial effects:

[0043] This invention first evaluates the stability of electrocardiogram (ECG) data based on differences and time intervals at different locations. Then, it applies mean filtering to each data point using time windows at different scales, obtaining data sequences at each scale. Next, based on the data changes around the maxima, the time intervals between adjacent maxima, and the correlation between the data sequences and respiratory rate data at each scale, combined with the periodic oscillation characteristics unique to heart failure, it accurately identifies heart failure specificity. Compared to single-scale analysis, this provides a more comprehensive and detailed reflection of the dynamic changes in cardiac function of the target individual, effectively enhancing the reference value of the features. Furthermore, it analyzes the heart failure specificity of the target individual at each time scale, determining error parameters by combining sample entropy from different data sequences. This allows for the extraction of parameters that capture core data change characteristics, reducing redundant information interference with the model, lowering computational complexity and overfitting risk, thereby achieving the determination of the target individual's level of attention and improving the accuracy and reliability of the determination results. Attached Figure Description

[0044] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating the method performed by a deep learning-based heart failure patient mortality risk assessment system provided in an embodiment of the present invention. Detailed Implementation

[0046] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description of the deep learning-based mortality risk assessment system for heart failure patients proposed according to the present invention is provided in conjunction with the accompanying drawings and preferred embodiments.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0048] The specific scheme of the deep learning-based heart failure patient mortality risk assessment system provided by the present invention will be described in detail below with reference to the accompanying drawings.

[0049] Example of a deep learning-based mortality risk assessment system for heart failure patients:

[0050] The specific scenario addressed in this embodiment is as follows: In the process of assessing the mortality risk of heart failure patients, it is necessary to monitor the target personnel. When all the monitored indicators show significant fluctuations or abnormalities, higher attention should be paid. Therefore, this embodiment will comprehensively analyze the data of the target personnel from different dimensions to determine whether abnormalities have occurred, and determine the level of attention based on the results.

[0051] This embodiment proposes a deep learning-based mortality risk assessment system for heart failure patients, such as... Figure 1 As shown, the method performed by the deep learning-based heart failure patient mortality risk assessment system in this embodiment includes the following steps:

[0052] Step S1: Obtain the electrocardiogram data and respiratory rate data of the target personnel.

[0053] First, the ECG and respiratory rate data of the target personnel within the current time period are obtained through the hospital information system. The current time period is the set of all historical moments with a time interval less than or equal to a preset duration. The preset duration is set by the implementer according to specific circumstances, and will not be elaborated further here. In this embodiment, the sampling frequency of both ECG and respiratory rate data is set to 125Hz. In specific applications, the implementer can set this according to specific circumstances. It should be noted that the data in this embodiment was obtained with full authorization. Then, a bandpass filter is used to filter the obtained ECG data to eliminate baseline drift. It should be noted that the ECG data mentioned below refers to the filtered data.

[0054] Thus, this embodiment has obtained the electrocardiogram data and respiratory rate data of the target personnel.

[0055] Step S2: Based on the data differences and time intervals at different locations on the electrocardiogram (ECG) data, the stability is obtained; mean filtering is performed on each data point on the ECG data using time windows of different scales to obtain data sequences at each scale; based on the data changes on both sides of the maximum point in the data sequence at each scale, the time interval between adjacent maximum points, the stability, and the correlation between the data sequence at each scale and the respiratory rate data, the heart failure cycle-specific values ​​of the data sequence at each scale are obtained.

[0056] By leveraging the unique advantage of long-term monitoring in the electrocardiogram data of target individuals, dynamic features are extracted and transformed into multi-scale rhythm instability features. These features are then integrated with static clinical data to provide a dynamic and low-cost enhancement dimension for their mortality risk assessment.

[0057] The absolute value of the difference between adjacent data points on the electrocardiogram (ECG) data is recorded as the first difference, meaning multiple first differences are obtained, with one first difference existing between every two adjacent data points. The ratio between the time interval between the maximum and minimum values ​​on the ECG data and the numerical difference between the maximum and minimum values ​​on the ECG data is calculated, and this ratio is recorded as the first ratio.

[0058] The stability is obtained based on the difference between each first difference and the average of all first differences and the first ratio. The difference between each first difference and the average of all first differences is negatively correlated with the stability, and the first ratio is positively correlated with the stability.

[0059] Among them, a positive correlation means that the dependent variable increases as the independent variable increases, and the dependent variable decreases as the independent variable decreases. It can be an additive relationship, a multiplicative relationship, etc., which is determined by practical application. A negative correlation means that the dependent variable decreases as the independent variable increases, and the dependent variable increases as the independent variable decreases. It can be a subtractive relationship, a division relationship, etc., which is determined by practical application.

[0060] In this embodiment, a formula for calculating the degree of stability is given, and the degree of stability can be specifically expressed as:

[0061]

[0062] in, Indicates the degree of stability. This indicates the number of data points on the electrocardiogram (ECG) data. This represents the data difference between the i-th data point and the (i+1)-th data point in the electrocardiogram data, also known as the first difference; This represents the average difference between all adjacent data points in an electrocardiogram (ECG) dataset. This represents the time interval between the maximum and minimum values ​​in electrocardiogram (ECG) data. This represents the numerical difference between the maximum and minimum values ​​in electrocardiogram (ECG) data. This indicates the preset first adjustment parameter. Indicates the absolute value sign. This represents the normalization function.

[0063] In this embodiment, a preset first adjustment parameter is introduced into the formula for calculating the stability level to prevent the denominator from being 0. In this embodiment, the preset first adjustment parameter is 0.01. In specific applications, the implementer can set it according to the specific situation. This represents the first ratio. The smaller the difference between adjacent data, the smaller the short-term fluctuations in the ECG data during the monitoring process, which means it is more stable. The larger the ratio is in terms of time scale, the more effective the feature can be captured. The larger the difference between extreme values ​​in the data sequence, the less stable the data sequence is. Furthermore, if the time interval between the maximum and minimum values ​​is smaller, the data sequence should have a smaller time scale to capture features that change significantly in the short term.

[0064] Mean filtering is applied to each data point in the electrocardiogram (ECG) data using preset time windows of different scales. The preset time window duration ranges from [30, 500], where the difference between two adjacent data points is 10. That is, the duration of the first scale time window is 30, the second scale time window is 40, the third scale time window is 50, and so on, with a 10-degree difference between adjacent scale time windows, resulting in multiple data sequences at different scales. In practical applications, the implementer can set the duration of different dimensions of the time window according to the specific situation. Mean filtering is an existing technology and will not be elaborated further here.

[0065] For data sequences across all scales, to accurately extract the rhythmic features of the target individuals, multi-scale entropy is used to quantify this rhythmic instability, and this instability is fused with static clinical features as a dynamic feature, thereby improving the performance of the deep learning-based mortality risk assessment model. It may exhibit rapid fluctuations on short timescales, while showing a slower trend on long timescales.

[0066] Calculate the sample entropy for each data sequence. Sample entropy is a measure of the time series complexity of the data sequence. The smaller the sample entropy, the lower the complexity of the data sequence and the higher the self-similarity. The larger the sample entropy, the more complex and variable the rhythm of the target person is, which is caused by healthy autonomic nervous system regulation. When the entropy value is larger, it indicates more abnormal data and late-stage decompensated heart failure. However, before calculating the sample entropy, it is necessary to perform preliminary identification of the data in the data sequence to include periodic oscillations unique to heart failure patients.

[0067] The more evenly distributed the maxima in a data sequence, the more similar the degree of adjacent rises and falls, and the more similar the intervals between maxima, the more stable its oscillations. Based on these characteristics, the oscillation performance of the data sequence at each scale can be determined.

[0068] For any scale:

[0069] Calculate the average slope of the data points on each side of each maximum point at this scale, and record this average slope as the first average slope value. The method for obtaining the maximum points is as follows: fit the corresponding fitted curve to all data curves in the data sequence at this scale, and extract the maximum points on the curve as the maximum points at this scale. It should be noted that when calculating the average slope of the data points on each side of each maximum point, for any given maximum point, the data points on each side are all the data points between it and its adjacent extreme points.

[0070] The absolute value of the difference between the first average slope values ​​corresponding to both sides of each maximum point at this scale is denoted as the second difference for each maximum point at this scale. The ratio between the number of maxima in the data sequence at this scale and the average time interval between all adjacent maxima points is calculated and denoted as the second ratio. Based on the second differences of all maxima points at this scale and the second ratio, the oscillation performance value of the data sequence at this scale is obtained. The second difference is negatively correlated with the oscillation performance value, and the second ratio is positively correlated with the oscillation performance value.

[0071] In this embodiment, a specific formula for calculating the oscillation performance value is given. The oscillation behavior of a data sequence at each scale can be expressed as:

[0072]

[0073] in, Indicates the first The oscillation performance of a data sequence at various scales. Indicates the first The number of maxima in a data sequence at each scale. This represents the average time interval between all adjacent maxima. Indicates the first The average of the second difference of all maxima at each scale. This indicates the preset second adjustment parameter. Indicates the absolute value sign.

[0074] No. The smaller the average of the second differences of all maxima at each scale, the more similar the ratio of the upward and downward slopes of the data on both sides of the maxima, and the more oscillating the changes in the data are. The more maxima in the data sequence, and the more similar the intervals between adjacent maxima, the more it can be used as data support for the periodic oscillation of the data sequence, and the higher its oscillation performance value.

[0075] In data sequences with smaller time scales, the respiratory signals of the target individual can affect the spacing between adjacent R waves on the electrocardiogram leads, forming RSA. Therefore, it is necessary to determine the correlation of respiratory signals at each scale. When a person breathes normally, there is a coupling correlation between respiratory signals and heart rate. However, for patients with heart failure, the coherence between the data sequence and respiration is significantly reduced. Furthermore, in data sequences with larger time scales, slow-cycle coupling occurs due to changes in respiratory patterns caused by heart failure. For example, multiple breaths may correspond to one heart rate fluctuation, which is an abnormal manifestation of the RR interval. The correlation of the respiratory RR interval is significantly reduced at smaller time scales. Therefore, the smaller the time scale, the less accurate the coupling results obtained.

[0076] The respiratory rate data sequence is obtained by sorting all the collected respiratory rate data in chronological order.

[0077] For any scale:

[0078] The Dynamic Time Warping (DTW) distance between the data sequence and the respiratory rate data sequence at this scale is calculated. The negative correlation normalized result of this DTW distance is taken as the correlation between the two sequences, and this correlation is recorded as the first correlation for this scale. The process of obtaining the negative correlation normalized result of the DTW distance is as follows: the value of an exponential function with the natural constant as the base and the negative DTW distance as the exponent is used as the negative correlation normalized result of the DTW distance. The scale corresponding to the highest correlation data is more effective in capturing the characteristics of heart failure patients. Thus, the respiratory correlation at each scale is determined. Each scale has a corresponding first correlation.

[0079] The absolute value of the difference between the length of the time window at this scale and the length of the time window at the scale corresponding to the largest first correlation is denoted as the third difference. The product of the third difference and the first correlation at this scale is determined as the respiratory correlation at this scale.

[0080] The stronger the oscillation value of the target person at this scale, and the weaker the respiratory correlation at this scale, the more disordered the data sequence data and the respiratory status are, the weaker the coupling is, and the worse the state of autonomic nervous system imbalance of the target person.

[0081] Based on the above characteristics, this embodiment combines oscillatory performance values, respiratory correlation, and stability to determine the heart failure cycle-specific value of the data sequence at this scale. The oscillatory performance value is positively correlated with the heart failure cycle-specific value, while the respiratory correlation and stability are both negatively correlated with the heart failure cycle-specific value.

[0082] In this embodiment, a specific method for obtaining the heart failure cycle-specific value is given, which is as follows: Calculate the sum of the respiratory correlation and the zero-prevention parameter corresponding to the scale, where the zero-prevention parameter is greater than 0. In this embodiment, the zero-prevention parameter is 0.01. In specific applications, the implementer can set it according to the specific situation. The ratio of the oscillation performance value of the target person at this scale to the sum is recorded as the third ratio. The product of the reciprocal of the stability and the third ratio is determined as the heart failure cycle-specific value of the data sequence at this scale.

[0083] Using the above method, heart failure cycle-specific values ​​of the data sequence at each scale can be obtained.

[0084] Step S3: Determine the error parameters of the data sequences at each scale based on the sample entropy of the data sequences at each scale, the difference between sample entropies at different scales, and the heart failure cycle-specific values.

[0085] For each scale of the data sequence, its sample entropy is obtained; the sample entropy reflects the complex rhythmicity of the RR interval of the data sequence. The higher the entropy value, the rich rhythmic changes of a healthy heart are corresponding to; the lower the entropy value, the more likely heart failure may occur due to sympathetic hyperactivity of the heart compressing normal fluctuations.

[0086] The sample entropy of the data sequences at all scales is sorted in ascending order to obtain the sample entropy sequence. Curve fitting is then performed on the data in the sample entropy sequence to obtain the sample entropy curve.

[0087] Next, based on the fluctuations of the sample entropy curve and combined with the specificity of the heart failure cycle, a reasonable parameter will be quantified for effective learning of the mortality risk assessment system for heart failure patients based on deep learning.

[0088] For each inflection point, the degree of scale shift is such that the larger the scale difference between each segment between inflection points and the greater the change in sample entropy, the better it reflects the unit scale change within that segment. Since the calculation of sample entropy only focuses on amplitude changes, and the periodic oscillations unique to heart failure are difficult to capture, the error parameters are determined by combining specific evaluation values ​​at different time scales.

[0089] For any scale: the ratio of the absolute value of the slope of the sample entropy curve at that scale to the sample entropy of the data sequence at that scale is denoted as the third ratio; the product of the third ratio and the heart failure cycle-specific value of the data sequence at that scale is determined as the error parameter of the data sequence at that scale. Using this method, the error parameter of the data sequence at each scale can be obtained.

[0090] Step S4: Use the error parameter and the sample entropy to determine the level of attention to the target personnel.

[0091] In multi-scale entropy analysis, changes in sample entropy at different time scales contain key information about heart failure data. To accurately extract indicators that reflect the true characteristics of the target individuals, error parameters at all time scales need to be systematically screened. The specific process is as follows: First, calculate the decrease in sample entropy between two adjacent time scales. This decrease value characterizes the degree of change in cardiac rhythm complexity over time. By sorting all decrease values, select the values ​​with the largest preset number of decrease values. In this embodiment, the preset number is 3. In specific applications, the implementer can set it according to the specific situation. This strategy is based on the following logic: the larger the decrease value, the more drastic the transition of cardiac rhythm from relatively complex (high entropy value) to significantly simpler (low entropy value) within the corresponding time scale change interval. This transition highly matches the abnormal rhythm regulation caused by autonomic nervous system dysfunction and myocardial electrical remodeling during the development of heart failure, and therefore best reflects the true characteristics of the target individuals. By selecting these three time scales with the largest decrease values ​​and extracting their corresponding error parameters as key parameters, the trend characteristics of cardiac rhythm at different time scales are preserved, while the sensitivity to heart failure-specific patterns is significantly enhanced. Ultimately, these key parameters serve as input features for the deep learning model, improving the accuracy and reliability of the model's output.

[0092] To comprehensively assess abnormal situations involving target personnel, this embodiment employs a deep learning model that fuses static and dynamic features, combined with edge computing and a risk warning mechanism for processing. The specific steps are as follows:

[0093] 1. Feature extraction and branching: A dual-branch structure is constructed. The static branch takes in clinical features (such as age, medical history, laboratory indicators, etc.) and extracts high-level semantic information through a 2-layer fully connected network (FCN). The dynamic branch takes 3 key features as input and captures the multi-scale dynamic changes of cardiac rhythm through a 1-layer FCN.

[0094] 2. Feature fusion and model training: The output vectors of the static and dynamic branches are concatenated and input into the fusion FCN layer to realize feature interaction. Finally, the anomaly index is output through the output layer, and historical data is used to train and optimize the model.

[0095] 3. Clinical Application and Risk Warning: Clinicians can use the output of the neural network to determine whether to give higher attention to a target individual, thereby assessing their mortality risk. The higher the abnormal indicator, the higher the attention should be given. When the abnormal indicator output by the neural network exceeds a preset abnormal threshold, the system immediately sends an alert to the doctor, prompting timely attention to the corresponding individual and appropriate intervention measures. The preset abnormal threshold is set by the implementer according to the specific situation, and will not be elaborated further here.

[0096] This completes the identification of the target individuals' status, allowing doctors to assess their mortality risk based on subsequent observations.

[0097] This embodiment first evaluates the stability of ECG data based on data differences and time intervals at different locations. Then, it uses time windows of different scales to perform mean filtering on each data point in the ECG data to obtain data sequences at each scale. Next, based on the data changes on both sides of the maximum points in the data sequences at each scale, the time interval between adjacent maximum points, and the correlation between the data sequences at each scale and respiratory rate data, combined with the periodic oscillation characteristics unique to heart failure, it accurately identifies the specificity of heart failure. Compared with single-scale analysis, it can more comprehensively and meticulously reflect the dynamic changes in the cardiac function of the target person, effectively improving the reference value of the features. Furthermore, it analyzes the specificity of heart failure in the target person at each time scale, and determines the error parameters by combining the sample entropy of different data sequences. This allows the parameters that extract the core data change characteristics to be extracted, reducing the interference of redundant information on the model, reducing computational complexity and the risk of overfitting, thereby realizing the determination of the target person's attention and improving the accuracy and reliability of the determination results.

[0098] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based mortality risk assessment system for heart failure patients, characterized in that, The system includes a memory and a processor, the processor executing a computer program stored in the memory to perform the following steps: Obtain electrocardiogram and respiratory rate data of the target personnel; Based on the data differences and time intervals at different locations on the electrocardiogram (ECG) data, the stationarity is obtained; mean filtering is performed on each data point on the ECG data using time windows of different scales to obtain data sequences at each scale; based on the data changes around the maxima in the data sequences at each scale, the time interval between adjacent maxima, the stationarity, and the correlation between the data sequences at each scale and respiratory rate data, the heart failure cycle-specific values ​​of the data sequences at each scale are obtained, including: for any scale: Based on the difference in the degree of data change between the left and right sides of each maximum point at any given scale, the number of maximum points, and the time interval between adjacent maximum points, the oscillation performance value of the data sequence at any given scale is obtained. The correlation between the data sequence at any scale and the respiratory rate data sequence is denoted as the first correlation corresponding to any scale. The respiratory rate data sequence is obtained by sorting the respiratory rate data in chronological order. The respiratory correlation degree corresponding to any of the first correlation scales is obtained; By combining the oscillatory performance value, the respiratory correlation, and the stability, the heart failure cycle specific value of the data sequence at any scale is determined. The oscillatory performance value is positively correlated with the heart failure cycle specific value, while the respiratory correlation and the stability are both negatively correlated with the heart failure cycle specific value. Based on the sample entropy of the data sequence at each scale, the difference between sample entropy at different scales, and the heart failure cycle-specific value, the error parameters of the data sequence at each scale are determined. The level of attention to the target personnel is determined using the error parameter and the sample entropy.

2. The deep learning-based mortality risk assessment system for heart failure patients according to claim 1, characterized in that, The method of obtaining stability based on data differences and time intervals at different locations on the electrocardiogram data includes: The difference between adjacent data points in the electrocardiogram data is recorded as the first difference. The stability is obtained based on the difference between each first difference and the average of all first differences, the time interval between the maximum and minimum values ​​on the electrocardiogram data, and the numerical difference between the maximum and minimum values ​​on the electrocardiogram data.

3. The deep learning-based mortality risk assessment system for heart failure patients according to claim 2, characterized in that, The method of obtaining the stability based on the difference between each first difference and the average of all first differences, the time interval between the maximum and minimum values ​​on the electrocardiogram data, and the numerical difference between the maximum and minimum values ​​on the electrocardiogram data includes: Calculate a first ratio between the time interval and the numerical difference; The stability is obtained based on the difference between each first difference and the average of all first differences and the first ratio. The difference between each first difference and the average of all first differences is negatively correlated with the stability, and the first ratio is positively correlated with the stability.

4. The deep learning-based mortality risk assessment system for heart failure patients according to claim 1, characterized in that, The step of obtaining the oscillation performance value of the data sequence at any scale based on the difference in the degree of data change on the left and right sides of each maximum point, the number of maximum points, and the time interval between adjacent maximum points includes: Calculate the first average slope value of each data point on each side of each maximum point at any given scale; The difference between the first average slope values ​​corresponding to both sides of each maximum point at any scale is recorded as the second difference of each maximum point at any scale. Calculate a second ratio between the number of maxima in the data sequence at any given scale and the average time interval between all adjacent maxima. Based on the second difference and the second ratio of all maxima at any given scale, the oscillation performance value of the data sequence at any given scale is obtained. The second difference is negatively correlated with the oscillation performance value, and the second ratio is positively correlated with the oscillation performance value.

5. The deep learning-based mortality risk assessment system for heart failure patients according to claim 1, characterized in that, The step of obtaining the respiratory correlation corresponding to any scale based on the first correlation includes: The difference between the length of the time window at any scale and the length of the time window at the scale corresponding to the largest first correlation is denoted as the third difference; The product of the third difference and the first correlation corresponding to any scale is determined as the respiratory correlation corresponding to any scale.

6. The deep learning-based mortality risk assessment system for heart failure patients according to claim 1, characterized in that, The determination of the heart failure cycle-specific value of the data sequence at any scale by combining the oscillatory performance value, the respiratory correlation, and the stability includes: Calculate the sum of the respiratory correlation coefficient and the zero-prevention parameter, where the zero-prevention parameter is greater than 0; Calculate the third ratio of the oscillation performance value to the sum value, and determine the product of the reciprocal of the stability degree and the third ratio as the heart failure cycle-specific value of the data sequence at any scale.

7. The deep learning-based mortality risk assessment system for heart failure patients according to claim 1, characterized in that, The step of determining the error parameters of the data sequence at each scale based on the sample entropy of the data sequence at each scale, the difference between sample entropies at different scales, and the heart failure cycle-specific value includes: Sort the sample entropy of the data sequences at all scales in ascending order to obtain the sample entropy sequence, and then perform curve fitting on the data in the sample entropy sequence to obtain the sample entropy curve. For any scale: Based on the slope value of the sample entropy curve of the given scale, the sample entropy of the data sequence at the given scale, and the heart failure cycle-specific value, the error parameters of the data sequence at the given scale are obtained.

8. The deep learning-based mortality risk assessment system for heart failure patients according to claim 7, characterized in that, The step of obtaining the error parameters of the data sequence at any scale based on the slope value of the sample entropy curve at any scale, the sample entropy of the data sequence at any scale, and the heart failure cycle-specific value includes: The ratio of the absolute value of the slope of the sample entropy curve at any scale to the sample entropy of the data sequence at any scale is denoted as the third ratio; the product of the third ratio and the heart failure cycle specificity of the data sequence at any scale is determined as the error parameter of the data sequence at any scale.

9. The deep learning-based mortality risk assessment system for heart failure patients according to claim 1, characterized in that, The process of determining the level of attention to the target personnel using the error parameter and the sample entropy includes: Sort all scales in descending order of sample entropy to obtain a scale sequence; use the error parameters of the data sequences at the first preset number of scales in the scale sequence as key parameters; The key parameters are input into the trained neural network to obtain predicted values; The level of attention given to the target personnel is determined based on the predicted values.

Images