Atrial fibrillation recurrence prediction method and system based on multi-modal data

Abstract

The present application relates to the technical field of medical data mining, in particular to a method and system for predicting recurrence of atrial fibrillation based on multi-modal data. The method divides the level period based on the historical symptom level duration; determines the atrial fibrillation risk index by combining the symptom level change and the keyword addition similarity in the level period; determines the normal period and filters out the problem period by the periodic fluctuation of the dynamic electrocardiogram; analyzes the problem performance degree according to the difference between the problem period and the normal period, the problem distribution of the level period and the risk index, and determines the prediction attention degree by referring to the frequent confusion of the previous problem distribution, and adjusts the prediction weight based on the prediction attention degree to make prediction and warning. The present application integrates multi-source data and dynamically quantifies the risk, adjusts the prediction weight of the high-risk period, effectively improves the individualization and accuracy of the prediction of recurrence of atrial fibrillation, and provides more reliable support for recurrence warning.

Description

Technical Field

[0001] This invention relates to the field of medical data mining technology, specifically to a method and system for predicting atrial fibrillation recurrence based on multimodal data. Background Technology

[0002] Atrial fibrillation (AF) is a common cardiac arrhythmia characterized by rapid, irregular beating of the atria. Its recurrence is insidious, sudden, and varies among individuals, and the risk of complications such as stroke and heart failure is significantly increased after recurrence. Therefore, predicting the risk of AF recurrence and intervening early is crucial to reducing adverse patient outcomes.

[0003] Since atrial fibrillation recurrence is a dynamic process, its occurrence is closely related to the patient's physiological state, clinical indicators, and treatment response at different time points. That is, during a certain period before atrial fibrillation recurrence, the fluctuations of certain indicators may be more significant than in other periods. For example, recurrence is more likely during the vagal nerve excitation period at night. Therefore, more attention needs to be paid to this period when predicting atrial fibrillation recurrence in order to better capture these dynamic changes and improve the predictive accuracy of the model.

[0004] However, due to individual patient differences, the physiological state and clinical manifestations of different patients may vary greatly at different times. When the attention given to different time periods is based solely on the magnitude of the fluctuation of the indicators, it often fails to reflect the individual differences of patients well. For example, the value of fluctuation data after exercise for recurrence prediction is not consistent for patients with different disease conditions. Fixed weights will mask this dynamic characteristic and may lead to inappropriate attention given to different time periods, thus reducing the accuracy of atrial fibrillation recurrence prediction. Summary of the Invention

[0005] To address the technical problems in the prior art, the present invention aims to provide a method and system for predicting atrial fibrillation recurrence based on multimodal data. The specific technical solution adopted is as follows:

[0006] This invention provides a method for predicting atrial fibrillation recurrence based on multimodal data, the method comprising:

[0007] Obtain the atrial fibrillation symptom level and diagnostic report of each patient's examination, and extract keywords from the diagnostic report; obtain the patient's dynamic electrocardiogram over time through wearable electrocardiogram device;

[0008] Based on the persistence of atrial fibrillation symptom levels from historical examinations, the time periods are divided into different levels. Within each level time period, atrial fibrillation risk indicators are determined based on the maintenance and changes in atrial fibrillation symptom levels, as well as the addition of similar keywords among the diagnostic reports.

[0009] Within each time period, the time period is divided based on the frequency cycle of the time-series dynamic electrocardiogram; normal time periods are determined from the time period based on the maximum similarity of the dynamic electrocardiograms between the time period periods; problem time periods are screened out based on the differences between the dynamic electrocardiograms of each abnormal time period and the local time period.

[0010] For each problem period, the degree of problem manifestation is analyzed based on the differences in dynamic electrocardiograms between the problem period and the normal period, as well as the distribution of problem periods and atrial fibrillation risk indicators in the problem period. The predictive attention level for each problem period is determined by combining the frequent and chaotic distribution of the preceding problem periods.

[0011] The predicted electrocardiogram is adjusted and an early warning is issued based on the predicted attention level during the problem cycle period.

[0012] Furthermore, the method for obtaining the atrial fibrillation risk indicators includes:

[0013] For any given grade period, the duration of that grade period is negatively correlated and normalized to obtain the time maintenance instability of that grade period. The difference in atrial fibrillation symptom level between that grade period and each adjacent grade period is calculated, and the mean of all atrial fibrillation symptom level differences is normalized to obtain the change maintenance instability of that grade period.

[0014] By combining the severity of atrial fibrillation symptoms, the degree of instability in maintaining the condition, and the degree of instability in maintaining the condition over time during that period, the severity assessment for that period can be obtained.

[0015] The set of keywords that did not appear in all previous checks during each grade period is taken as the new set for each check; for any two checks during the grade period, the set of keywords that are different between the two checks is taken as the different set of the two checks.

[0016] The proportion of the number of keywords appearing in the different sets in each of the two checks is used as the different new symptom indicators for each check; the mean values ​​of the different new symptom indicators in the two checks are negatively correlated and combined with the intersection-union ratio of the keyword sets in the two checks to obtain the new similarity indicators for the two checks.

[0017] The sum of all newly added similar indicators from every two examinations within this grade period is negatively correlated and used as an indicator for assessing symptom complexity within this grade period.

[0018] By combining the severity assessment and symptom complexity assessment indicators for this grade of time period, the risk index for atrial fibrillation in this grade of time period is obtained.

[0019] Furthermore, the method for obtaining the periodic time segment includes:

[0020] The dynamic electrocardiogram of each level time period is converted to the frequency domain space. The reciprocal of the frequency corresponding to the first peak in the frequency domain space is used as the period of each level time period. Each level time period is divided according to the corresponding period to obtain the periodic time period.

[0021] Furthermore, the method for determining the normal time period includes:

[0022] Within each grade period, the similarity of dynamic electrocardiograms between periodic periods is used as a metric to cluster the periodic periods, thus obtaining periodic clusters;

[0023] The periodic cluster with the most periodic time intervals is designated as the normal cluster, and the periodic time intervals in the normal cluster are designated as normal time intervals.

[0024] Furthermore, the method for obtaining the problem time period includes:

[0025] For any abnormal time period, calculate the similarity between the abnormal time period and the dynamic electrocardiogram of each other time period within the preset local range, and take the mean of the similarity of all dynamic electrocardiograms as the local stability of the abnormal time period.

[0026] Abnormal periods with local stability below a preset threshold are designated as problem periods.

[0027] Furthermore, the method for obtaining the problem performance degree includes:

[0028] For any problem time period, calculate the similarity between the problem time period and each normal time period in the dynamic electrocardiogram, and perform negative correlation mapping on the mean of all dynamic electrocardiogram similarities to obtain the normal deviation index of the problem time period.

[0029] The ratio of the number of all problem periods in the time period of the problem period to the total number of all period periods is calculated. Combined with the atrial fibrillation risk index of the time period of the problem period, the data is normalized to obtain the overall risk level of the problem period.

[0030] By combining the overall risk level and normal deviation index of the problem period, the problem performance level of the problem period is obtained.

[0031] Furthermore, the method for obtaining the predicted attention level includes:

[0032] For any problem period, the period consisting of the problem period and a preset number of consecutive period periods in the time sequence is used as the preceding analysis period of the problem period.

[0033] The ratio of the number of problematic time periods in the preceding analysis period to the total number of time periods in the total cycle is used as the frequency indicator of the problematic time period; the similarity of the dynamic electrocardiograms between every two problematic time periods in the preceding analysis period is calculated, and the mean of all similarities in the preceding analysis period is negatively correlated to obtain the problem confusion index.

[0034] Obtain the interval duration between every two adjacent problem periods in the preceding analysis period; perform negative correlation mapping on the mean of all interval durations to obtain the preceding frequency index of the problem period;

[0035] By combining the frequency index, the problem disorder index, and the preceding frequent index for this problem period, the problem frequency for this problem period is obtained;

[0036] By combining the frequency and performance of the problem, the prediction adjustment coefficient for the problem period is obtained. The prediction adjustment coefficient is a normalized value. The sum of the prediction adjustment coefficient for the problem period and the value of 1 is taken as the predicted attention level for the problem period.

[0037] Furthermore, the method of weighting and adjusting the predicted electrocardiogram based on the predicted attention level during the problem period and issuing an early warning includes:

[0038] Based on the predicted attention during the problem period, the dynamic electrocardiogram of the problem period is weighted, and the weighted ARIMA model is used to predict the patient's electrocardiogram signal to obtain the predicted signal;

[0039] When the similarity between the predicted signal and the electrocardiogram during historical atrial fibrillation exceeds a preset similarity threshold, an alert is issued.

[0040] Furthermore, the method for obtaining the graded time period includes:

[0041] Tests with equal and continuous distribution of atrial fibrillation symptoms in time sequence are considered as graded distribution tests, and the time period corresponding to each graded distribution test is considered as each graded time period; the time corresponding to the first test in the graded distribution test is considered as the start time of the graded time period, and the time corresponding to the last test in the graded distribution test is considered as the end time of the graded time period.

[0042] The present invention also provides an atrial fibrillation recurrence prediction system based on multimodal data, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the atrial fibrillation recurrence prediction method based on multimodal data as described above.

[0043] The present invention has the following beneficial effects:

[0044] This invention combines diagnostic data with monitored dynamic electrocardiogram (ECG) data to more comprehensively identify the historical characteristics of atrial fibrillation recurrence in patients through multimodal data. By dividing the patient's condition into time periods based on the persistence of historical symptom levels, the changes in the patient's condition can be broken down into stage units, allowing risk analysis to focus on specific time periods and improving the specificity of the analysis. Within each time period, atrial fibrillation risk indicators are determined by combining changes in symptom level maintenance with similarities in newly added keywords, quantifying the risk differences between different time periods. Dynamic symptom changes more accurately reflect the risk of the condition, making risk assessment more aligned with the individual patient's situation. Further analysis from a physiological perspective identifies normal time periods by analyzing the periodic fluctuations of the dynamic ECG and filters out problematic time periods that may indicate atrial fibrillation instability, excluding normal fluctuations unrelated to atrial fibrillation such as exercise, thus improving the accuracy of identifying abnormal time periods. For problematic time periods, the severity of the problem is analyzed by combining the differences from normal time periods, the distribution of problems within the specific time level, and risk indicators. The predictive focus is determined by referencing the frequency and irregularity of previous problem distributions. Weights are determined by comprehensively considering the frequency of normal differences within the problematic time period and the overall severity of the time level, highlighting the importance of high-risk periods and assigning them higher weight in subsequent predictions to improve the accuracy of atrial fibrillation recurrence prediction. This invention effectively enhances the personalization and accuracy of atrial fibrillation recurrence prediction through the integration of multi-source data and dynamic risk quantification, providing more reliable support for recurrence early warning. Attached Figure Description

[0045] 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.

[0046] Figure 1 A flowchart of a method for predicting atrial fibrillation recurrence based on multimodal data, provided in one embodiment of the present invention;

[0047] Figure 2 This is a partial schematic diagram of a dynamic electrocardiogram provided in one embodiment of the present invention;

[0048] Figure 3 A flowchart illustrating a method for obtaining atrial fibrillation risk indicators according to an embodiment of the present invention;

[0049] Figure 4 This is a flowchart of a method for obtaining predicted attention levels according to an embodiment of the present invention. Detailed Implementation

[0050] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a method and system for predicting atrial fibrillation recurrence based on multimodal data proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0051] 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.

[0052] The following description, in conjunction with the accompanying drawings, details the specific scheme of the atrial fibrillation recurrence prediction method and system based on multimodal data provided by this invention.

[0053] Please see Figure 1 The diagram illustrates a flowchart of a method for predicting atrial fibrillation recurrence based on multimodal data, according to an embodiment of the present invention. The method includes the following steps:

[0054] S1: Obtain the atrial fibrillation symptom level and diagnostic report for each patient's examination, and extract keywords from the diagnostic report; obtain the patient's dynamic electrocardiogram over time using a wearable ECG device.

[0055] For any atrial fibrillation patient, in this embodiment of the invention, relevant data from each examination is retrieved from a hospital database. This includes data such as electrocardiogram (ECG), echocardiogram, blood tests, and clinical manifestations. Based on the examination results, the doctor writes a detailed diagnostic report for each examination, with each report having a corresponding timestamp. Each test item has a normal reference range, and the report indicates whether the patient's test results are within the normal range. If certain indicators exceed the normal range, the report typically briefly explains the possible causes or significance. For example, abnormal TSH levels may indicate hyperthyroidism or hypothyroidism, electrolyte imbalances may affect cardiac electrical activity, and coagulation abnormalities may require adjustment of anticoagulation therapy.

[0056] When atrial fibrillation recurs, patients may experience symptoms such as palpitations, shortness of breath, and fatigue. Doctors typically inquire about changes in these symptoms and record them in the diagnostic report. To facilitate subsequent analysis of the characteristic symptom descriptions in the diagnostic report, in this embodiment of the invention, for each diagnostic report, a Rapid Automatic Keyword Extraction (Rake) algorithm is used to extract keywords by analyzing stop words and phrase boundaries in the report, resulting in a keyword set. It should be noted that keyword extraction technology is a well-known technique among those skilled in the art and will not be limited or elaborated upon here.

[0057] Each examination yields the atrial fibrillation symptom level for the patient at that time. In one specific embodiment of the invention, the atrial fibrillation symptom level is determined according to the European Heart Rhythm Association (EHRA) recommended symptom level, used to assess the severity of the patient's symptoms and their impact on daily life. Specific grading includes: Grade 1 (Asymptomatic): Atrial fibrillation does not cause any symptoms, assigned a grade of 1 in this embodiment. Grade 2a (Mild Symptoms): Daily activities are not affected by atrial fibrillation-related symptoms, but the patient may feel distressed by the symptoms, assigned a grade of 2 in this embodiment. Grade 2b (Moderate Symptoms): Daily activities are not affected by atrial fibrillation-related symptoms, but the patient is bothered by the symptoms, assigned a grade of 3 in this embodiment. Grade 3 (Severe Symptoms): Daily activities are significantly affected by atrial fibrillation-related symptoms, assigned a grade of 4 in this embodiment. Grade 4 (Disabling Symptoms): Daily activities are completely stopped due to atrial fibrillation-related symptoms, assigned a grade of 5 in this embodiment.

[0058] It should be noted that this scoring system is widely used in clinical practice and research to help doctors assess the severity of symptoms in patients with atrial fibrillation, and will not be elaborated further here.

[0059] For patients with intermittent or asymptomatic atrial fibrillation, 24-hour or longer Holter monitoring is required to detect recurrences of atrial fibrillation. Wearable ECG devices can be used for long-term monitoring, which can help detect recurrences of atrial fibrillation in daily life. Please refer to... Figure 2 The diagram shows a partial schematic of a dynamic electrocardiogram provided in an embodiment of the present invention.

[0060] S2: Based on the persistence of atrial fibrillation symptom levels from historical examinations, classify the time periods into different levels; within each time period, determine the atrial fibrillation risk indicators for each time period based on the maintenance and changes in atrial fibrillation symptom levels and the addition of similar keywords among the diagnostic reports from the examinations.

[0061] The core trigger for atrial fibrillation recurrence is instability or worsening of the condition, and patient test data is the direct basis for reflecting changes in the condition. However, a single grade rating cannot reflect temporal differences in risk. For example, a patient may rapidly progress from asymptomatic Grade 1 to Grade 3 with limited daily activities within 3 months. The risk of Grade 3 within 3 months is clearly higher than that of Grade 1 lasting for 1 year. Because atrial fibrillation does not change at a uniform rate, but has periods of stability and periods of worsening, the risk of recurrence differs significantly between different stages. Therefore, the historical timeline is initially divided based on the persistence of atrial fibrillation symptom grades, with consecutive occurrences of the same grade reflecting a period of stable condition.

[0062] In this embodiment of the invention, the method for dividing the time period for obtaining the grade includes:

[0063] Tests with equal and continuous atrial fibrillation symptom levels over time are considered as tests with the same level distribution. The time period corresponding to each test with the same level distribution is considered as each level time period. For example, if the atrial fibrillation symptom level distribution over time is 1112211333, then the atrial fibrillation symptom levels corresponding to each test with the same level distribution are 111, 22, 11, and 333.

[0064] The time corresponding to the first inspection in the same level distribution inspection is taken as the start time of the level period, and the time corresponding to the last inspection in the same level distribution inspection is taken as the end time of the level period. In this embodiment of the invention, the time corresponding to the inspection is the time corresponding to the timestamp when the diagnostic report is generated in the inspection.

[0065] The essence of time period segmentation is to transform discrete test results into continuous time period risk assessment. By segmenting time periods into levels, more attention is paid to the disease trend and symptom stability within each stable time period, thereby assessing the degree to which the disease is more likely to relapse during that time period.

[0066] Even within the same stage, patients' symptoms may vary significantly. For example, some patients may have stable symptoms, while others may develop multiple new symptoms. New symptoms often indicate a more complex condition, such as impaired cardiac function. Furthermore, the recurrence of atrial fibrillation may be closely related to cardiac structural abnormalities, electrophysiological changes, underlying diseases such as hypertension, coronary heart disease, and diabetes, as well as lifestyle factors such as smoking, alcohol consumption, and obesity.

[0067] Therefore, grading alone cannot reflect this implicit risk. While considering the potential adjustment of atrial fibrillation severity analysis based on the continuous temporal changes within each grade stage, it is also necessary to consider the impact of complex changes in characteristic symptom keywords in the diagnostic report. Patients with more complex symptom changes within the same grade have a higher risk of relapse. For example, in a grade 2 period, there may be two examinations. The first examination only shows palpitations, while the second examination adds shortness of breath and chest pain. This symptom complexity may indicate disease progression, such as worsening myocardial ischemia. In contrast, the risk of persistent palpitations alone in a grade period is relatively low.

[0068] Therefore, preferably, in this embodiment of the invention, the method for obtaining atrial fibrillation risk indicators for different time periods by combining the changes and maintenance of the grade period with the consistency of keyword additions is described in the following reference. Figure 3 The diagram illustrates a flowchart of a method for obtaining atrial fibrillation risk indicators according to an embodiment of the present invention. The method includes the following steps:

[0069] S201: Adjustments are made based on the inter-adjacent changes and persistence of the grade period to obtain the severity assessment of the grade period.

[0070] Since a single severity level cannot reflect the dynamic severity of risk over time, combining the duration of severity assessment with the degree of change between adjacent severity levels is more effective. The greater the degree of change between a severity level and adjacent severity levels over time, and the shorter the duration, the worse the stability of the severity level is, and the higher the risk of relapse. Therefore, dynamically adjusting the severity assessment level is more effective.

[0071] In this embodiment of the invention, for any given time period, the value obtained by negatively correlating and normalizing the duration of that time period is used as the time stability of that time period; the shorter the duration, the worse the stability. It should be noted that negative correlation mapping and normalization are techniques well-known to those skilled in the art. Negative correlation mapping can take the form of a negative exponential power or an inverse proportional form, and normalization can be linear normalization or standard normalization, etc. The methods are not elaborated upon or limited here.

[0072] Further calculations were performed to determine the difference in atrial fibrillation symptom levels between the current grade period and each adjacent grade period. The mean of all atrial fibrillation symptom level differences was normalized to obtain the stability of the change in the grade period. The higher the difference in atrial fibrillation symptom levels between the grade period and adjacent grade periods in time sequence, the greater the grade fluctuation, reflecting poorer symptom stability and a higher risk of unstable recurrence.

[0073] Finally, by combining the atrial fibrillation symptom severity, variability instability, and temporal instability of the corresponding time period, the severity assessment score for that time period is obtained. The atrial fibrillation symptom severity, variability instability, and temporal instability are all positively correlated with the severity assessment score. Higher scores in these three parameters reflect a more severe atrial fibrillation condition and poorer dynamic temporal stability, thus requiring a higher severity assessment. In this embodiment of the invention, the product of the atrial fibrillation symptom severity, variability instability, and temporal instability of the corresponding time period is used as the severity assessment score for that time period.

[0074] By analyzing the distribution trend of grade changes over time, the long-term changes in a patient's condition are broken down into quantifiable stage characteristics. This improves the accuracy of the analysis and assessment of different stages of disease stability and deterioration from the perspective of changes over time, and avoids masking the characteristics of high-risk periods by only looking at the course of the disease through grade assessment.

[0075] S202: Obtain symptom complexity assessment indicators for each graded time period by assessing the complexity of newly added keywords between examinations.

[0076] At different grade levels, newly added symptoms often represent early signs of accumulated cardiac function. Even if the grade remains unchanged, more complex symptoms in the diagnostic report reflect a more significant pre-relapse warning sign. Furthermore, considering that some patients may experience symptom worsening even without a grade increase due to increased tolerance, it's crucial to consider the addition of key symptoms in each examination at different grade levels. The lower the overlap between examinations at the same grade level, the more new symptoms appear at that grade level, indicating more complex symptom changes and a situation where the grade remains stable but the actual risk is increased. In such cases, a higher severity assessment should be applied to the corresponding grade level.

[0077] In this embodiment of the invention, the method for obtaining symptom complexity assessment indicators by adding complex keywords includes:

[0078] First, the set of keywords that did not appear in all previous checks during each grade period is taken as the new set for each check. That is, the intersection of the keywords of each check and all previous checks in the time sequence is removed from the keyword set of each check to obtain the new set. This new set represents the keywords that do not belong to all historical checks in the keyword set of each check, and is the new keyword symptom that appears in the current check.

[0079] For any two examinations within this grade period, the set of keywords that differ between the two examinations is taken as the set of differences between the two examinations, reflecting the different symptoms that exist between the two examinations at the same grade.

[0080] Furthermore, the proportion of keywords appearing in the new set within the different sets in each of the two examinations is used as a different indicator of newly emerging symptoms for each examination. That is, the ratio of the number of keywords in the intersection of the new set and the different sets to the number of keywords in the different sets is used as a different indicator of newly emerging symptoms. A larger indicator indicates that a large number of new disease keywords, never before seen, have appeared in the different disease keywords in the two examinations, meaning the patient's condition has undergone new changes, resulting in more complex impacts. In such cases, the similarity analysis needs to be adjusted downwards.

[0081] Then, the mean values ​​of the newly added different symptom indicators from the two examinations are negatively correlated and combined with the intersection-union ratio (CIU) of the keyword sets from the two examinations to obtain the newly added similarity indicators from the two examinations. The CIU measures the degree of the same symptoms between the two examinations. The larger the CIU, the higher the similarity between the sets. At this time, the larger the combined newly added different symptom indicators from the two examinations, the less reliable the similarity.

[0082] In this embodiment of the invention, the mean values ​​of the newly added different symptom indicators from the two examinations are negatively correlated and normalized, and then multiplied by the intersection-union ratio of the keyword sets of the two examinations to obtain the newly added similarity indicators of the two examinations.

[0083] Finally, the sum of the newly added similarity indicators from every two examinations within the grade period was negatively correlated and used as an indicator to assess the complexity of symptoms within that grade period. The worse the newly added similarity between each pair of examinations within the comprehensive grade period, the worse the overall symptom similarity after adjustment through the newly added analysis, the more complex the atrial fibrillation phenomenon is, which is related to multiple factors and therefore requires more attention.

[0084] By introducing symptom similarity analysis, it can be found that, under the same level of illness, the addition or change of symptoms often indicates an increase in the complexity of the illness, which is an important basis for judging whether the risk has been implicitly aggravated, thus reducing the limitations of the analysis.

[0085] S203: Combine the severity assessment and symptom complexity assessment indicators of this grade period to obtain the atrial fibrillation risk index for this grade period.

[0086] Finally, by comprehensively assessing the severity level and symptom changes, the risk of atrial fibrillation (AF) is evaluated for each level period. Both the severity assessment score and the symptom complexity assessment index are positively correlated with AF risk indicators. Higher scores indicate more complex symptom changes and greater risk within that level period, requiring closer attention to this data in AF recurrence prediction. In this embodiment, the product of the severity assessment score and the symptom complexity assessment index for that level period is used as the AF risk indicator for that level period.

[0087] S3: Within each grade period, divide the periodic time into time-series periods based on the frequency cycle of the dynamic electrocardiogram; determine the normal period from the time-series period based on the maximum similarity of the dynamic electrocardiograms between the time-series periods; and filter out the problem period based on the differences between the dynamic electrocardiograms of each abnormal period and the local period.

[0088] After analyzing the results of regular checkups, due to the gap between adjacent checkups, real-time monitoring with Holter monitoring is necessary. This can detect the presence, frequency, and duration of atrial fibrillation, which is the main method for diagnosing atrial fibrillation. Recurrence of atrial fibrillation is often accompanied by sudden abnormalities in real-time ECG signals, such as a sudden and irregular increase in heart rate. However, fluctuations in ECG signals can be caused by normal factors such as exercise and emotions, requiring differentiation between atrial fibrillation-related abnormalities and physiological fluctuations.

[0089] Under normal circumstances, the heart rhythm fluctuates periodically. Therefore, the periodicity of the dynamic electrocardiogram (ECG) is used to divide the time frame into periodic segments for comparative analysis. In this embodiment of the invention, the dynamic ECG at each level is converted to the frequency domain. The reciprocal of the frequency corresponding to the first peak in the frequency domain is used as the period of each level. The level time frames are divided according to the corresponding periods to obtain multiple periodic time frames.

[0090] In one specific embodiment of the present invention, a short-time Fourier transform algorithm is used to convert the electrocardiogram (ECG) signal into a signal in the frequency domain to analyze its frequency distribution. In the frequency domain, the main frequency components of the ECG signal, such as the heart rate frequency component, can be observed. For an ECG signal, the fundamental frequency usually corresponds to the heart rate. By analyzing the peak values ​​in the spectrum, the fundamental frequency of the ECG signal can be determined. The fundamental frequency is usually the first significant peak in the spectrum. Therefore, the reciprocal of the fundamental frequency is denoted as the period of the ECG signal. For example, when the fundamental frequency is 1 Hz, the period is 1 second. It should be noted that frequency domain transformation and peak value determination are well-known techniques to those skilled in the art and will not be further elaborated here.

[0091] Since normal heart rhythms have stable cycles and high signal similarity, the periodic time segment of the normal heart rhythm segment can be determined by the maximum similarity. In this embodiment of the invention, within each level period, the dynamic electrocardiogram similarity between periodic time segments is used as a metric to cluster the periodic time segments, obtaining periodic clusters. In a specific embodiment of the invention, the DTW algorithm is used to obtain the DTW distance between dynamic electrocardiogram signals as a clustering metric. The clustering algorithm can be the K-means clustering algorithm. It should be noted that the smaller the DTW distance, the more similar the signals. The DTW algorithm and clustering algorithm are well-known techniques to those skilled in the art and will not be elaborated here.

[0092] Due to electrocardiogram Figure 1Generally, the periodic clusters are in a normal state most of the time. Therefore, the periodic clusters with the most periodic time periods are considered normal clusters, and the periodic time periods in the normal clusters are considered normal time periods. Periodic time periods other than normal time periods are considered time periods with fluctuations.

[0093] While normal activities such as exercise, eating, and emotional fluctuations can cause significant differences in electrocardiogram (ECG) signals compared to normal conditions, these do not lead to arrhythmia. Furthermore, changes in ECG signals are gradual and follow a certain pattern, meaning there are usually no sudden, large differences between adjacent time periods. Atrial fibrillation, on the other hand, is characterized by rapid, irregular beatings of the atria. This means that while the ECG signal differs significantly from normal conditions, the heart rate changes suddenly and irregularly, indicating substantial differences in ECG signals within adjacent time periods.

[0094] Therefore, in this embodiment of the invention, for deviations in abnormal time periods through local time periods, the method for filtering problem time periods includes:

[0095] For any abnormal time period, the similarity between the abnormal time period and each other cycle time period within a preset local range is calculated. The mean of the similarity of all dynamic electrocardiograms is taken as the local stability of the abnormal time period. In this embodiment of the invention, the preset local range is set to the size of two cycle time periods to the left and right of the abnormal time period. The specific range setting can be adjusted by the implementer according to the specific implementation scenario. The similarity calculation of electrocardiogram signals can be obtained by negative correlation mapping through DTW distance. In other embodiments of the invention, Manhattan distance can also be used. Signal similarity calculation is a technical means well known to those skilled in the art and is not limited here.

[0096] By using a threshold determination, abnormal periods with local stability less than a preset threshold are identified as problem periods. The greater the local stability, the higher the similarity of the local electrocardiogram signal, and the more likely it is to be a normal fluctuation. Otherwise, it may be an abnormal cycle of atrial fibrillation. In this embodiment of the invention, the preset threshold is set to 0.6, and the implementer can adjust the specific value as needed.

[0097] The system divides real-time electrocardiogram signals into periods and identifies abnormalities, eliminating interference from physiological fluctuations and focusing more on the analysis of actual abnormal periods.

[0098] S4: For each problem period, analyze the degree of problem manifestation based on the difference between the dynamic electrocardiogram and the normal period, the distribution of problem periods in the corresponding grade period, and the risk indicators of atrial fibrillation. Combine this with the frequent and chaotic distribution of the preceding problem periods to determine the predictive attention level for each problem period.

[0099] By further subdividing the cycle time periods, weights are assigned to the abnormality characteristics of the rhythm cycle. When atrial fibrillation recurrence prediction occurs during a period of frequent and concentrated problem cycles, and the more irregular and chaotic the heart rate is, the more attention should be paid to the signal characteristics of such periods to ensure that the prediction model focuses on key risk signals.

[0100] Therefore, the severity of the arrhythmia abnormalities and the frequency of localized occurrences of the problem during each problem period are used to assess the level of attention required for each problem period. Preferably, in this embodiment of the invention, the method for obtaining the predicted level of attention is described in [reference needed]. Figure 4 The diagram illustrates a flowchart of a method for predicting attention acquisition according to an embodiment of the present invention, which includes the following steps:

[0101] S401: For any problem time period, the problem manifestation degree is obtained based on the degree of difference between the problem time period and the normal time period in the dynamic electrocardiogram, as well as the distribution of problem time periods in the time period of the same level and atrial fibrillation risk indicators.

[0102] For each problematic time period, the greater the difference between the ECG signal during the problematic time period and the normal time period, the more significant the arrhythmia problem exhibited in this problematic time period, and the more attention it requires. Simultaneously, the analysis should also consider the grade of the problematic time period. If the grade of the time period contains more problems and the analyzed atrial fibrillation risk indicators are higher, then the risk abnormalities in each problematic time period within that grade are more significant and require greater attention.

[0103] Therefore, in this embodiment of the invention, the similarity between the problem period and each normal period of the dynamic electrocardiogram is calculated, and the mean of the similarity of all dynamic electrocardiograms is negatively correlated to obtain the normal deviation index of the problem period. When the similarity is smaller, the overall deviation is larger, and the electrocardiogram difference is greater.

[0104] Further analysis is performed on the ratio of all problematic time periods within the specified timeframe to the total number of all timeframes. This ratio is then normalized using the atrial fibrillation risk index for the specified timeframe to obtain the overall risk level of the problematic timeframe. A higher ratio reflects a higher degree of problem within the specified timeframe. In this embodiment, the ratio of all problematic time periods within the specified timeframe to the total number of all timeframes is multiplied by the atrial fibrillation risk index and then normalized to obtain the overall risk level, reflecting the problem performance within the specified timeframe.

[0105] Finally, by combining the overall risk performance level and the normal deviation index of the problem period, the problem performance level of the problem period is obtained. In this embodiment of the invention, the product of the overall risk performance level and the normal deviation index of the problem period is taken as the problem performance level of the problem period. The higher the problem performance level, the more attention the problem period needs to receive.

[0106] S402: Obtain the frequency of the problem based on the frequency and disorder of the problem time distribution in the preceding time sequence.

[0107] Since atrial fibrillation recurrence is often not an isolated, single abnormality, it usually manifests as paroxysmal, that is, frequent and chaotic arrhythmias within a short period of time, which can terminate spontaneously or be persistent, that is, require medical intervention to terminate. Therefore, when a patient's heart rate problems occur more frequently and chaotically, the abnormality is more sensitive, and the problematic period can be given greater weight.

[0108] Therefore, in this embodiment of the invention, for any problem time period, the time period consisting of the problem time period and a preset number of consecutive period periods preceding it is used as the preceding analysis time period of the problem time period. The frequency and disorder of the problem time period are analyzed on the preceding time period of the problem time period. The preset number can be set to 30, and the implementer can adjust it according to the specific implementation scenario.

[0109] Furthermore, the ratio of the number of problematic periods in the preceding analysis period to the total number of periods in the total cycle is used as a frequency indicator for the problematic period. The larger the frequency indicator, the higher the frequency of the problematic period in the short term.

[0110] Then, the similarity of the dynamic electrocardiogram between every two problem periods in the preceding analysis period is calculated. The mean of all similarities in the preceding analysis period is negatively correlated to obtain the problem disorder index. The lower the similarity between the electrocardiogram signals when each problem period occurs, the more disordered the heart rhythm is in the short term.

[0111] Furthermore, the interval between each two adjacent problem periods in the preceding analysis period is obtained, and the mean of all intervals is negatively correlated to obtain the preceding frequency index of the problem period. The shorter the interval, the more concentrated the frequency time and the more dense the problem periods.

[0112] Therefore, by combining the frequency index, problem confusion index, and preceding frequency index of the problem period, the problem frequency of the problem period can be obtained. The frequency index, problem confusion index, and preceding frequency index are all positively correlated with the problem frequency. The larger the frequency index, problem confusion index, and preceding frequency index, the more concentrated and frequent the problem period is and the more irregular the heart rhythm. This problem period needs to be given more attention.

[0113] S403: Based on the frequency and performance of the problem, obtain the predicted attention level for the problem during the specified time period.

[0114] The frequency and performance of a problem reflect the level of attention that can be given to a problem period. In this embodiment of the invention, the prediction adjustment coefficient for the problem period is obtained by combining the frequency and performance of the problem. The prediction adjustment coefficient is a normalized value. In this embodiment of the invention, the product of the frequency and performance of the problem is normalized and used as the prediction adjustment coefficient.

[0115] Finally, the sum of the prediction adjustment coefficient for the problem period and the value 1 is taken as the predicted attention level for that problem period. At this time, the predicted attention level for other non-problem periods is assumed to be the value 1.

[0116] S5: Adjust the predicted electrocardiogram based on the predicted attention level during the problem cycle period and issue an early warning.

[0117] Therefore, by identifying high-risk periods for atrial fibrillation recurrence and assigning them greater predictive weight, the overall prediction accuracy can be improved. By comparing accurate prediction data with historical data, atrial fibrillation recurrence warnings can be issued, reminding patients in advance and ensuring that they seek medical attention in a timely manner.

[0118] By predicting attention levels, the prediction weight of problem periods can be improved. In this embodiment of the invention, the dynamic electrocardiogram of the problem period is weighted based on the predicted attention levels of the problem period. A weighted ARIMA model is used to predict the patient's electrocardiogram signal to obtain the predicted signal. By obtaining the electrocardiogram signal of the next day, the similarity between it and the historical atrial fibrillation electrocardiogram signal can be further analyzed to predict recurrence in advance.

[0119] When the similarity between the predicted signal and the electrocardiogram during historical atrial fibrillation exceeds a preset similarity threshold, it indicates a high probability of atrial fibrillation recurrence, and a warning is issued. In one specific embodiment of the present invention, the preset similarity threshold can be set to 0.5, and the implementer can adjust the specific value as needed. It should be noted that the prediction model is a technique well-known to those skilled in the art and will not be described in detail here.

[0120] In summary, this invention combines diagnostic information with monitored dynamic electrocardiogram (ECG) data to more comprehensively identify the historical characteristics of atrial fibrillation recurrence in patients through multimodal data. By dividing the patient's condition into time periods based on the persistence of historical symptom levels, the changes in the patient's condition can be broken down into phased units, allowing risk analysis to focus on specific time periods and improving the specificity of the analysis. Within each time period, combining changes in symptom level maintenance with similarities in newly added keywords determines atrial fibrillation risk indicators, quantifying the risk differences at different time periods. Dynamic symptom changes more accurately reflect the risk of the condition, making risk assessment more aligned with the individual patient's situation. Furthermore, from a physiological perspective, the periodic fluctuations of the dynamic ECG identify normal time periods and screen for problematic time periods that may indicate atrial fibrillation instability, excluding normal fluctuations unrelated to atrial fibrillation such as exercise, thus improving the accuracy of identifying abnormal time periods. For problematic time periods, the severity of the problem is analyzed by combining the differences from normal time periods, the distribution of problems within the specific time level, and risk indicators. The predictive focus is determined by referencing the frequency and irregularity of previous problem distributions. Weights are determined by comprehensively considering the frequency of normal differences within the problematic time period and the overall severity of the time level, highlighting the importance of high-risk periods and assigning them higher weight in subsequent predictions to improve the accuracy of atrial fibrillation recurrence prediction. This invention effectively enhances the personalization and accuracy of atrial fibrillation recurrence prediction through the integration of multi-source data and dynamic risk quantification, providing more reliable support for recurrence early warning.

[0121] The present invention also provides an atrial fibrillation recurrence prediction system based on multimodal data, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the atrial fibrillation recurrence prediction method based on multimodal data as described above.

[0122] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0123] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for predicting atrial fibrillation recurrence based on multi-modal data, the method comprising: The method includes: Obtain the atrial fibrillation symptom level and diagnostic report of each patient's examination, and extract keywords from the diagnostic report; obtain the patient's dynamic electrocardiogram over time through wearable electrocardiogram device; Based on the persistence of atrial fibrillation symptom levels from historical examinations, the time periods are divided into different levels. Within each level time period, atrial fibrillation risk indicators are determined based on the maintenance and changes in atrial fibrillation symptom levels, as well as the addition of similar keywords among the diagnostic reports. Within each time period, the time period is divided based on the frequency cycle of the time-series dynamic electrocardiogram; normal time periods are determined from the time period based on the maximum similarity of the dynamic electrocardiograms between the time period periods; problem time periods are screened out based on the differences between the dynamic electrocardiograms of each abnormal time period and the local time period. For each problem period, the degree of problem manifestation is analyzed based on the differences in dynamic electrocardiograms between the problem period and the normal period, as well as the distribution of problem periods and atrial fibrillation risk indicators in the problem period. The predictive attention level for each problem period is determined by combining the frequent and chaotic distribution of the preceding problem periods. Based on the predicted attention level during the problem cycle, the predicted electrocardiogram is weighted and adjusted for early warning; The methods for obtaining the atrial fibrillation risk indicators include: For any given grade period, the duration of that grade period is negatively correlated and normalized to obtain the time maintenance instability of that grade period. The difference in atrial fibrillation symptom level between that grade period and each adjacent grade period is calculated, and the mean of all atrial fibrillation symptom level differences is normalized to obtain the change maintenance instability of that grade period. By combining the severity of atrial fibrillation symptoms, the degree of instability in maintaining the condition, and the degree of instability in maintaining the condition over time during that period, the severity assessment for that period can be obtained. The set of keywords that did not appear in all previous checks during each grade period is taken as the new set for each check; for any two checks during the grade period, the set of keywords that are different between the two checks is taken as the different set of the two checks. The proportion of the number of keywords appearing in the different sets in each of the two checks is used as the different new symptom indicators for each check; the mean values ​​of the different new symptom indicators in the two checks are negatively correlated and combined with the intersection-union ratio of the keyword sets in the two checks to obtain the new similarity indicators for the two checks. The sum of all newly added similar indicators from every two examinations within this grade period is negatively correlated and used as an indicator for assessing symptom complexity within this grade period. By combining the severity assessment and symptom complexity assessment indicators for this grade of time period, the risk index for atrial fibrillation in this grade of time period is obtained.

2. The method of claim 1, wherein the method is based on multi-modal data. The method for obtaining the periodic time period includes: The dynamic electrocardiogram of each level time period is converted to the frequency domain space. The reciprocal of the frequency corresponding to the first peak in the frequency domain space is used as the period of each level time period. Each level time period is divided according to the corresponding period to obtain the periodic time period.

3. The method of claim 1, wherein the method is based on multi-modal data. The method for determining the normal time period includes: Within each grade period, the similarity of dynamic electrocardiograms between periodic periods is used as a metric to cluster the periodic periods, thus obtaining periodic clusters; The periodic cluster with the most periodic time intervals is designated as the normal cluster, and the periodic time intervals in the normal cluster are designated as normal time intervals.

4. The method of claim 1, wherein the method is based on multi-modal data. The methods for obtaining the problem time period include: For any abnormal time period, calculate the similarity between the abnormal time period and the dynamic electrocardiogram of each other time period within the preset local range, and take the mean of the similarity of all dynamic electrocardiograms as the local stability of the abnormal time period. Abnormal periods with local stability below a preset threshold are designated as problem periods.

5. The method of claim 1, wherein the method is based on multi-modal data. The methods for obtaining the problem performance include: For any problem time period, calculate the similarity between the problem time period and each normal time period in the dynamic electrocardiogram, and perform negative correlation mapping on the mean of all dynamic electrocardiogram similarities to obtain the normal deviation index of the problem time period. The ratio of the number of all problem periods in the time period of the problem period to the total number of all period periods is calculated. Combined with the atrial fibrillation risk index of the time period of the problem period, the data is normalized to obtain the overall risk level of the problem period. By combining the overall risk level and normal deviation index of the problem period, the problem performance level of the problem period is obtained.

6. The method of predicting atrial fibrillation recurrence based on multi-modal data according to claim 1, wherein, The method for obtaining the predicted attention includes: For any problem period, the period consisting of the problem period and a preset number of consecutive period periods in the time sequence is used as the preceding analysis period of the problem period. The ratio of the number of problematic time periods in the preceding analysis period to the total number of time periods in the total cycle is used as the frequency indicator of the problematic time period; the similarity of the dynamic electrocardiograms between every two problematic time periods in the preceding analysis period is calculated, and the mean of all similarities in the preceding analysis period is negatively correlated to obtain the problem confusion index. Obtain the interval duration between every two adjacent problem periods in the preceding analysis period; perform negative correlation mapping on the mean of all interval durations to obtain the preceding frequency index of the problem period; By combining the frequency index, the problem disorder index, and the preceding frequent index for this problem period, the problem frequency for this problem period is obtained; By combining the frequency and performance of the problem, the prediction adjustment coefficient for the problem period is obtained. The prediction adjustment coefficient is a normalized value. The sum of the prediction adjustment coefficient for the problem period and the value of 1 is taken as the predicted attention level for the problem period.

7. The method of claim 1, wherein the method is based on multi-modal data. The method of weighting and adjusting the predicted electrocardiogram based on the predicted attention level during the problem cycle period and issuing an early warning includes: Based on the predicted attention during the problem period, the dynamic electrocardiogram of the problem period is weighted, and the weighted ARIMA model is used to predict the patient's electrocardiogram signal to obtain the predicted signal; When the similarity between the predicted signal and the electrocardiogram during historical atrial fibrillation exceeds a preset similarity threshold, an alert is issued.

8. The method of claim 1, wherein the method is based on multi-modal data. The methods for obtaining the graded time period include: Tests with equal and continuous distribution of atrial fibrillation symptoms in time sequence are considered as graded distribution tests, and the time period corresponding to each graded distribution test is considered as each graded time period; the time corresponding to the first test in the graded distribution test is considered as the start time of the graded time period, and the time corresponding to the last test in the graded distribution test is considered as the end time of the graded time period.

9. An atrial fibrillation recurrence prediction system based on multi-modal data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor implements the steps of the atrial fibrillation recurrence prediction method based on multi-modal data according to any one of claims 1-8 when executing the computer program.

Images