HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology

Abstract

The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, which belongs to the field of heart failure prediction, including: collecting electrocardiograms of different patients, and automatically identifying and classifying the subtle changes of each electrocardiogram based on the electrocardiogram learning algorithm to obtain the corresponding early prediction results and early detection accuracy; integrating the physiological data of each patient to establish a heart assessment association with its own electrocardiogram, and combining with the corresponding early detection accuracy to obtain an optimization vector; optimizing the electrocardiogram learning algorithm based on all optimization vectors to obtain an optimized learning algorithm; continuing to analyze each electrocardiogram according to the optimized learning algorithm to obtain an analysis result, and combining with the corresponding early detection result to obtain the HFimpEF prediction result of the corresponding patient. Reduce the misjudgment that may be caused by a single signal and improve the prediction accuracy of HFimpEF in the later stage.

Description

Technical Field

[0001] The present invention relates to the technical field of heart failure prediction, and in particular to a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology. Background Art

[0002] Heart failure with improved ejection fraction (HFimpEF) is a form of heart failure characterized by a near-normal cardiac ejection fraction, but patients still show typical symptoms of heart failure. The current diagnosis of HFimpEF mainly relies on imaging techniques, such as echocardiography and MRI. Although these imaging methods can provide detailed information on cardiac structure and function, they have certain limitations in actual use. First, existing technologies such as echocardiography and MRI rely on the resolution and imaging quality of the equipment. For complex types of heart failure such as HFimpEF, imaging methods are difficult to provide accurate diagnostic information in the early stages of pathophysiological changes. This dependence limits the application scenarios of the technology and cannot be widely used in primary care or remote diagnosis. In addition, although traditional imaging techniques can evaluate the ejection fraction of the heart, they are not sensitive enough to the early pathophysiological changes of HFimpEF patients, and are prone to miss the best time for early diagnosis. Effective diagnosis often requires patients to have obvious symptoms before they can be made. In addition, these methods require patients to go to specific medical institutions for examination, which increases the economic and time costs of testing and cannot be monitored in real time and for a long time. Moreover, existing technologies lack effective predictions of the future progression of heart failure, especially in the early stages before heart function deteriorates significantly. Existing imaging technologies cannot accurately predict the course of a patient's condition. Therefore, existing technologies are limited in their effectiveness in identifying which patients may be at risk of heart failure with improved ejection fraction.

[0003] Therefore, the present invention proposes a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology. Summary of the invention

[0004] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, which automatically identifies subtle changes based on deep learning algorithms and electrocardiograms, effectively reduces dependence on medical institutions and optimizes the model in combination with physiological data, reduces misjudgments that may be caused by a single signal, and improves the prediction accuracy of HFimpEF in the later stage.

[0005] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, comprising:

[0006] Step 1: Collect ECGs from different patients, and automatically identify and classify the subtle changes in each ECG based on the ECG learning algorithm to obtain the corresponding early prediction results and early detection accuracy;

[0007] Step 2: Integrate each patient's physiological data to establish a cardiac assessment association with their own electrocardiogram, and combine it with the corresponding early detection accuracy to obtain an optimization vector;

[0008] Step 3: Optimizing the electrocardiogram learning algorithm based on all optimization vectors to obtain an optimized learning algorithm;

[0009] Step 4: Continue to analyze each electrocardiogram according to the optimization learning algorithm to obtain analysis results, and combine the corresponding early detection results to obtain the HFimpEF prediction results of the corresponding patients.

[0010] Preferably, before automatically identifying and classifying the subtle changes of each electrocardiogram based on the electrocardiogram learning algorithm, the method includes: constructing an electrocardiogram learning algorithm, specifically including:

[0011] Constructing an algorithm cluster, wherein the algorithm cluster includes a plurality of initial algorithms, and each initial algorithm is constructed by using a different neural network and trained by using the same electrocardiogram samples;

[0012] Set several interval graphs of slight changes in heart rate, and start each initial algorithm to determine the recognition state of each interval graph, and construct a state array;

[0013] Obtain the theoretical result matrix of each initial algorithm from the algorithm-parameter database;

[0014] According to the theoretical result matrix of each initial algorithm, the actual recognition row vector for each interval graph and the known parameter column vector, the recognition inaccurate parameters for the corresponding interval graph are obtained;

[0015] The inaccurate parameters of the identification based on all interval graphs constitute the influence vector of the corresponding initial algorithm;

[0016] Perform minimum impact analysis and maximum impact analysis on the state array and influence vector of the same initial algorithm to obtain the state change under different influence analyses;

[0017] A reliable verification coefficient is obtained according to the state change amount and in combination with the state array, and an initial algorithm corresponding to the maximum coefficient is extracted from all the reliable verification coefficients as an electrocardiogram recognition algorithm.

[0018] Preferably, the state change amounts under different impact analyses are obtained respectively, including:

[0019] Obtaining respectively from the parameter-feature comparison table a tiny feature set matching each influencing parameter in the influencing vector;

[0020] According to the manually labeled feature results of each heart rate micro-change interval diagram, and combined with the micro-feature set, the change interval is set for each influencing parameter ;

[0021]

[0022] in, represents the setting variation interval of the i1th influencing parameter; Represents the tiny feature set of the i1th influencing parameter Extracted features based on influencing parameters from the manually annotated feature results of the j1th interval graph Similarity function of ; Indicates the number of interval graphs; express The number of 0; represents the unit change amount matched with the i1th influencing parameter obtained from the parameter-unit change comparison table; Indicates all The minimum value in ; Indicates all The maximum value in ; Indicates all The variance of Indicates all The variance of

[0023] Based on the result of the manually marked features, it is determined whether the corresponding interval graph has a tiny feature that matches the influencing parameter. If so, each influencing parameter is adjusted toward the maximum according to the set variation interval to obtain the maximum influencing value;

[0024]

[0025] in, represents the maximum influence value of the i1th influencing parameter; ln represents the sign of the logarithmic function; e is a constant with a value of 2.7;

[0026] At the same time, each influencing parameter is adjusted to a smaller value according to the set variation interval to obtain a minimum influencing value;

[0027]

[0028] in, represents the minimum influence value of the i1th influencing parameter;

[0029] If it does not exist, set the value of the corresponding influencing parameter to 0;

[0030] The values ​​of the influencing parameters involved in each interval diagram are assigned to the influence vector, and combined with the state array to determine the corresponding state change, wherein the value assignment includes: assigning the maximum influence value of the influencing parameter and the value of 0 to the influence vector, and assigning the minimum influence value of the influencing parameter and the value of 0 to the influence vector.

[0031] Preferably, combining the state array to determine the corresponding state change amount includes:

[0032] Based on the assigned value of each influencing variable in the corresponding influencing vector, a first overall sum sum1 of the corresponding vector is calculated, and at the same time, a second overall sum sum2 of the value of each influencing variable in the corresponding influencing vector before adjustment is obtained;

[0033] According to the first whole sum sum1, the second whole sum sum2 and the number of 0 values ​​in the corresponding vector , determine the amount of state change;

[0034]

[0035] in, Indicates the corresponding state change; Indicates the number of 0 values ​​assigned in the corresponding vector; It represents the variance of the absolute value of the difference between each influencing variable before and after adjustment in the corresponding vector; Indicates the number of influencing variables set by the corresponding vector.

[0036] Preferably, the physiological data includes: blood pressure and blood oxygen of the corresponding patient.

[0037] Preferably, a cardiac assessment correlation with the own electrocardiogram is established, including:

[0038] Collect multiple sets of physiological data of the corresponding patients and draw fluctuation curves of each physiological parameter;

[0039] The electrocardiogram and all fluctuation curves are sequentially input into an evaluation model associated with the heart to obtain a heart evaluation association.

[0040] Preferably, obtaining the optimization vector includes:

[0041] Extracting the correlation between the physiological parameters and the electrocardiogram according to the correlation index of the heart assessment correlation, and then determining the influence factor of the corresponding correlation index on the corresponding early detection accuracy;

[0042] According to the influencing factors of each associated indicator, an optimization vector is constructed.

[0043] Preferably, the HFimpEF prediction result of the corresponding patient is obtained in combination with the corresponding early detection result, including:

[0044] Continuing to analyze each electrocardiogram based on the optimization learning algorithm to obtain an analysis result;

[0045] The analysis results of each electrocardiogram are comprehensively compared with the early detection results to obtain N1 pair arrays of each analysis parameter, where each pair array contains the early parameter value of the corresponding electrocardiogram under the same analysis parameter and the current parameter value after algorithm optimization, and N1 represents the number of electrocardiograms;

[0046] Draw a first curve of early parameter values ​​and a second curve of current parameter values ​​for the N1 array of pairs, and determine the intersection point between the first curve and the second curve;

[0047] If the number of intersection points is 0 or 1, the current parameter value of the corresponding analysis parameter will be retained;

[0048] If the number of intersection points is not 0 and is not 1, the first curve and the second curve are vertically divided according to the intersection points to sequentially obtain the curve area difference between every two intersection points;

[0049] According to the area differences of all involved curves, the current parameter values ​​of the corresponding analysis parameters are adjusted to obtain new values ​​and retain them;

[0050]

[0051] in, Indicates the corresponding adjusted new value; Indicates the corresponding current parameter value; And m1 represents the number of intersection points, and m2 represents the number of segmented curve segments based on the existence of the intersection points; represents the curve area difference of the j2-th segmented curve segment; represents the curve area difference of the j3rd segment, and ; It represents the number of all curve area differences greater than 0; It indicates the number of all curve area differences less than 0;

[0052] The retained values ​​of different analysis parameters under the corresponding electrocardiogram are used as the HFimpEF prediction results of the corresponding patients.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] It uses deep learning algorithms and electrocardiograms as a basis to automatically identify subtle changes, effectively reducing dependence on medical institutions and optimizing the model in combination with physiological data, reducing misjudgments that may be caused by a single signal and improving the prediction accuracy of HFimpEF in the later stage.

[0055] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 The figure is a flow chart of a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0060] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, such as Figure 1 As shown, including:

[0061] Step 1: Collect ECGs from different patients, and automatically identify and classify the subtle changes in each ECG based on the ECG learning algorithm to obtain the corresponding early prediction results and early detection accuracy;

[0062] Step 2: Integrate each patient's physiological data to establish a cardiac assessment association with their own electrocardiogram, and combine it with the corresponding early detection accuracy to obtain an optimization vector;

[0063] Step 3: Optimizing the electrocardiogram learning algorithm based on all optimization vectors to obtain an optimized learning algorithm;

[0064] Step 4: Continue to analyze each electrocardiogram according to the optimization learning algorithm to obtain analysis results, and combine the corresponding early detection results to obtain the HFimpEF prediction results of the corresponding patients.

[0065] In this embodiment, the subtle change refers to the change of the heart rate in the electrocardiogram.

[0066] In this embodiment, the early prediction result is the recognition and classification result of the electrocardiogram by the electrocardiogram learning algorithm, and the early detection accuracy is the detection accuracy of the electrocardiogram learning algorithm, both of which can be directly obtained.

[0067] In this embodiment, the physiological data refers to blood pressure data and blood oxygen data.

[0068] In this embodiment, the heart-related assessment refers to the impact of blood pressure and blood oxygen on heart rate.

[0069] In this embodiment, the optimization vector={the impact factor of each associated indicator}.

[0070] In this embodiment, the optimization process refers to the optimization vector continuing to learn the algorithm, and the algorithm after the learning is completed is the optimized learning algorithm.

[0071] In this embodiment, the analysis result is the current prediction result of the electrocardiogram by the optimization learning algorithm, that is, the existing heart condition result.

[0072] In this embodiment, the HFimpEF prediction result refers to a result obtained by auxiliary optimization of the analysis result based on the early detection result.

[0073] The beneficial effects of the above technical solution are: automatically identifying subtle changes based on deep learning algorithms and electrocardiograms, effectively reducing dependence on medical institutions and optimizing the model in combination with physiological data, reducing misjudgments that may be caused by a single signal, and improving the prediction accuracy of HFimpEF in the later stage.

[0074] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology. Before automatically identifying and classifying the subtle changes of each electrocardiogram based on an electrocardiogram learning algorithm, the method comprises: constructing an electrocardiogram learning algorithm, specifically comprising:

[0075] Constructing an algorithm cluster, wherein the algorithm cluster includes a plurality of initial algorithms, and each initial algorithm is constructed by using a different neural network and trained by using the same electrocardiogram samples;

[0076] Set several interval graphs of slight changes in heart rate, and start each initial algorithm to determine the recognition state of each interval graph, and construct a state array;

[0077] Obtain the theoretical result matrix of each initial algorithm from the algorithm-parameter database;

[0078] According to the theoretical result matrix of each initial algorithm, the actual recognition row vector for each interval graph and the known parameter column vector, the recognition inaccurate parameters for the corresponding interval graph are obtained;

[0079] The inaccurate parameters of the identification based on all interval graphs constitute the influence vector of the corresponding initial algorithm;

[0080] Perform minimum impact analysis and maximum impact analysis on the state array and influence vector of the same initial algorithm to obtain the state change under different influence analyses;

[0081] A reliable verification coefficient is obtained according to the state change amount and in combination with the state array, and an initial algorithm corresponding to the maximum coefficient is extracted from all the reliable verification coefficients as an electrocardiogram recognition algorithm.

[0082] In this embodiment, the neural network may be a convolutional neural network, a recurrent neural network, etc., and may be an initial algorithm obtained by combining one or more networks, all of which are existing algorithms.

[0083] In this embodiment, the interval graph refers to a known graph containing slight changes in heart rate, and is a known sample set for reasonable screening of the algorithm.

[0084] In this embodiment, the recognition state refers to the heart rate recognition result of each interval graph by the initial algorithm, so as to determine the heart function.

[0085] In this embodiment, the state array = {the recognition state of each interval graph by the same initial algorithm}.

[0086] In this embodiment, the algorithm-parameter database includes different initial algorithms and theoretical recognition results after the electrocardiogram is recognized based on the initial algorithms, that is, the theoretical recognition matrix .

[0087] In this embodiment, the actually recognized row vector is the result of the initial algorithm actually recognizing the interval graph during the actual application process.

[0088] In this embodiment, the row vector in the theoretical result matrix is ​​subtracted from the same element of the corresponding actual recognition row vector and the absolute value is taken. The known parameters whose absolute values ​​are greater than the set threshold of the corresponding column are screened, and then the screened known parameters are used as the recognition inaccurate parameters of the corresponding interval diagram. Each recognition result contains several parameters, which are all pre-set, such as parameters for the corresponding algorithm to recognize heart rate, parameters for the corresponding algorithm to recognize signal stability, parameters for the corresponding algorithm to recognize whether there is interference in the signal, etc.

[0089] In this embodiment, the influence vector refers to the result obtained by performing union processing on the inaccurate parameters of all interval graphs, that is, {various identification inaccurate parameters}.

[0090] In this embodiment, the reliability verification coefficient=((the state change amount under the maximum impact analysis+the state change amount under the minimum impact analysis) / 2)×(the number of states in the state array that match the theoretical heart rate function situation / the number of all electrocardiograms).

[0091] The beneficial effects of the above technical solution are: by constructing an algorithm cluster to identify different interval graphs and based on this to judge the actual recognition results of the analysis algorithm for different graphs and then lock the inaccurate parameters of the algorithm, further determine the state change by performing maximum and minimum analysis on the influence vector, and finally calculate the reliable verification coefficient to screen the final algorithm, providing an algorithm basis for subsequent HFimpEF prediction.

[0092] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, which obtains the state change under different influence analysis, including:

[0093] Obtaining respectively from the parameter-feature comparison table a tiny feature set matching each influencing parameter in the influencing vector;

[0094] According to the manually labeled feature results of each heart rate micro-change interval diagram, and combined with the micro-feature set, the change interval is set for each influencing parameter ;

[0095]

[0096] in, represents the setting variation interval of the i1th influencing parameter; Represents the tiny feature set of the i1th influencing parameter Extracted features based on influencing parameters from the manually annotated feature results of the j1th interval graph Similarity function of ; Indicates the number of interval graphs; express The number of 0; represents the unit change amount matched with the i1th influencing parameter obtained from the parameter-unit change comparison table; Indicates all The minimum value in ; Indicates all The maximum value in ; Indicates all The variance of Indicates all The variance of

[0097] Based on the result of the manually marked features, it is determined whether the corresponding interval graph has a tiny feature that matches the influencing parameter. If so, each influencing parameter is adjusted toward the maximum according to the set variation interval to obtain the maximum influencing value;

[0098]

[0099] in, represents the maximum influence value of the i1th influencing parameter; ln represents the sign of the logarithmic function; e is a constant with a value of 2.7;

[0100] At the same time, each influencing parameter is adjusted to a smaller value according to the set variation interval to obtain a minimum influencing value;

[0101]

[0102] in, represents the minimum influence value of the i1th influencing parameter;

[0103] If it does not exist, set the value of the corresponding influencing parameter to 0;

[0104] The values ​​of the influencing parameters involved in each interval diagram are assigned to the influence vector, and combined with the state array to determine the corresponding state change, wherein the value assignment includes: assigning the maximum influence value of the influencing parameter and the value of 0 to the influence vector, and assigning the minimum influence value of the influencing parameter and the value of 0 to the influence vector.

[0105] In this embodiment, the parameter-feature comparison table contains micro-feature sets of different parameters, which are pre-compared and stored, and the micro-feature set refers to micro-features of the electrocardiogram, such as atrial P' wave, retrograde P (P⁻) wave, atrial flutter wave (F wave), atrial fibrillation wave (f wave), etc., which are also set in advance.

[0106] In this embodiment, the manually labeled feature results are the results of expert labeling and have reference value.

[0107] In this embodiment, the influencing parameter refers to an inaccurate recognition parameter corresponding to the algorithm.

[0108] In this embodiment, the setting of the change interval is achieved according to the matching conditions of the features.

[0109] In this embodiment, the specific description of the influence vector is: {the value of influence parameter 1 the value of influence parameter 2 the value of influence parameter 2 ...}.

[0110] The beneficial effect of the above technical solution is: based on the matching analysis of the tiny feature set and the manually labeled feature results, the set change interval of each influencing parameter can be determined, and then the value of the parameter can be adjusted, providing a basis for the subsequent determination of the state change amount.

[0111] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, combined with a state array, to determine the corresponding state change, including:

[0112] Based on the assigned value of each influencing variable in the corresponding influencing vector, a first overall sum sum1 of the corresponding vector is calculated, and at the same time, a second overall sum sum2 of the value of each influencing variable in the corresponding influencing vector before adjustment is obtained;

[0113] According to the first whole sum sum1, the second whole sum sum2 and the number of 0 values ​​in the corresponding vector , determine the amount of state change;

[0114]

[0115] in, Indicates the corresponding state change; Indicates the number of 0 values ​​assigned in the corresponding vector; It represents the variance of the absolute value of the difference between each influencing variable before and after adjustment in the corresponding vector; Indicates the number of influencing variables set by the corresponding vector.

[0116] In this embodiment, the first overall sum sum1 includes two cases, one is the sum of the vectors adjusted to be smaller, and the other is the sum of the vectors adjusted to be larger, and each case has a corresponding state change amount.

[0117] The beneficial effect of the above technical solution is: by obtaining the overall sum before and after the adjustment of the influencing variable, the state change amount in the corresponding adjustment direction is comprehensively determined, providing a basis for the subsequent calculation of the reliability verification coefficient.

[0118] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, which establishes a cardiac assessment association with the electrocardiogram of the patient, including:

[0119] Collect multiple sets of physiological data of the corresponding patients and draw fluctuation curves of each physiological parameter;

[0120] The electrocardiogram and all fluctuation curves are sequentially input into an evaluation model associated with the heart to obtain a heart evaluation association.

[0121] Preferably, the physiological data includes: blood pressure and blood oxygen of the corresponding patient.

[0122] In this embodiment, the multiple sets of physiological data refer to blood pressure and blood oxygen at different times, and the fluctuation curve refers to a curve for blood pressure and a curve for blood oxygen.

[0123] In this embodiment, the heart-related evaluation model is obtained by training the neural network model based on different combinations of electrocardiograms, blood pressure curves, and blood oxygen curves and the heart function association results of the combinations as samples, that is, there is a certain correlation between blood pressure, blood oxygen and heart rate.

[0124] The beneficial effect of the above technical solution is: by obtaining the fluctuation curves of blood pressure and blood oxygen, and combining them with the electrocardiogram, the model performs heart-related evaluation, thereby ensuring the accuracy of the results.

[0125] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology to obtain an optimization vector, including:

[0126] Extracting the correlation between the physiological parameters and the electrocardiogram according to the correlation index of the heart assessment correlation, and then determining the influence factor of the corresponding correlation index on the corresponding early detection accuracy;

[0127] According to the influencing factors of each associated indicator, an optimization vector is constructed.

[0128] In this embodiment, the correlation index refers to the correlation between blood oxygen and electrocardiogram, the correlation between blood pressure and electrocardiogram, and the correlation between blood pressure and blood oxygen and electrocardiogram. Starting from these three aspects, the correlation can be extracted, and the correlation includes: strong correlation, weak correlation, and no correlation.

[0129] In this embodiment, the impact factor=the relevance of each correlation indicator×the algorithm learning rate of the ECG learning algorithm.

[0130] In this embodiment, the optimization vector={the impact factor of each associated indicator}.

[0131] The beneficial effect of the above technical solution is: the correlation under different indicators is extracted by cardiac correlation evaluation, and then the influencing factors are determined, which facilitates the algorithm optimization.

[0132] The present invention provides a HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, which combines the corresponding early detection results to obtain the HFimpEF prediction results of the corresponding patients, including:

[0133] Continuing to analyze each electrocardiogram based on the optimization learning algorithm to obtain an analysis result;

[0134] The analysis results of each electrocardiogram are comprehensively compared with the early detection results to obtain N1 pair arrays of each analysis parameter, where each pair array contains the early parameter value of the corresponding electrocardiogram under the same analysis parameter and the current parameter value after algorithm optimization, and N1 represents the number of electrocardiograms;

[0135] Draw a first curve of early parameter values ​​and a second curve of current parameter values ​​for the N1 array of pairs, and determine the intersection point between the first curve and the second curve;

[0136] If the number of intersection points is 0 or 1, the current parameter value of the corresponding analysis parameter will be retained;

[0137] If the number of intersection points is not 0 and is not 1, the first curve and the second curve are vertically divided according to the intersection points to sequentially obtain the curve area difference between every two intersection points;

[0138] According to the area differences of all involved curves, the current parameter values ​​of the corresponding analysis parameters are adjusted to obtain new values ​​and retain them;

[0139]

[0140] in, Indicates the corresponding adjusted new value; Indicates the corresponding current parameter value; And m1 represents the number of intersection points, and m2 represents the number of segmented curve segments based on the existence of the intersection points; represents the curve area difference of the j2-th segmented curve segment; represents the curve area difference of the j3rd segment, and ; It represents the number of all curve area differences greater than 0; It indicates the number of all curve area differences less than 0;

[0141] The retained values ​​of different analysis parameters under the corresponding electrocardiogram are used as the HFimpEF prediction results of the corresponding patients.

[0142] In this embodiment, the analysis result is the recognition result of the electrocardiogram by the optimized algorithm.

[0143] In this embodiment, the N1 pair array includes parameter values ​​of the same analysis parameter involved in N1 electrocardiograms, and the analysis parameters include characteristic recognition result parameters of the electrocardiogram, such as heart rate parameters, signal stability parameters, cardiac function parameters, etc.

[0144] In this embodiment, the first curve drawn is composed of the parameter values ​​of the corresponding analysis parameters obtained after each electrocardiogram is identified using the algorithm before optimization, and the second curve drawn is composed of the parameter values ​​of the corresponding analysis parameters obtained after each electrocardiogram is identified using the algorithm after optimization.

[0145] In this embodiment, the comparison of the recognition results of the electrocardiogram for the same parameter by the optimized algorithm is effectively determined by performing intersection analysis on the curves.

[0146] In this embodiment, if the number of intersection points is 0 or 1, the current parameter value of the corresponding analysis parameter is retained, which means that the current parameter values ​​of the corresponding analysis parameters of all electrocardiograms are kept unchanged.

[0147] In this embodiment, vertical segmentation refers to vertical division with the intersection point as the cutting point.

[0148] In this embodiment, the curve area difference refers to the difference between a second area of ​​the second curve based on two intersection points and a first area of ​​the first curve based on two intersection points.

[0149] In this embodiment, the curve area difference is positive or negative, so Let's analyze it further.

[0150] The beneficial effect of the above technical solution is: the electrocardiogram is identified and analyzed through the optimized algorithm, and the analysis results are comprehensively compared with the early detection results to obtain N1 pair arrays for each parameter, and then the number of intersection points is judged by curve drawing, and the judgment results in different situations are analyzed and adjusted in turn to ensure the accuracy of the final prediction results.

[0151] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology, characterized in that: include: Step 1: Collect electrocardiograms of different patients, and automatically identify and classify the subtle changes in each electrocardiogram based on the electrocardiogram learning algorithm to obtain the corresponding early detection results and early detection accuracy; Step 2: Integrate each patient's physiological data to establish a cardiac assessment association with their own electrocardiogram, and combine it with the corresponding early detection accuracy to obtain an optimization vector; Among them, establishing a cardiac assessment correlation with one's own electrocardiogram includes: Collect multiple sets of physiological data of the corresponding patients and draw fluctuation curves of each physiological parameter; Inputting the electrocardiogram and all fluctuation curves into an evaluation model associated with the heart in sequence to obtain a heart evaluation association; Step 3: Optimizing the electrocardiogram learning algorithm based on all optimization vectors to obtain an optimized learning algorithm; Step 4: Continue to analyze each electrocardiogram according to the optimization learning algorithm to obtain analysis results, and combine the corresponding early detection results to obtain the HFimpEF prediction results of the corresponding patients; The optimization vector is obtained, including: Extracting the correlation between the physiological parameters and the electrocardiogram according to the correlation index of the heart assessment correlation, and then determining the influence factor of the corresponding correlation index on the corresponding early detection accuracy; Construct an optimization vector based on the influencing factors of each associated indicator; Among them, combined with the corresponding early detection results, the HFimpEF prediction results of the corresponding patients are obtained, including: The analysis results of each electrocardiogram are comprehensively compared with the early detection results to obtain N1 pair arrays of each analysis parameter, where each pair array contains the early parameter value of the corresponding electrocardiogram under the same analysis parameter and the current parameter value after algorithm optimization, and N1 represents the number of electrocardiograms; Draw a first curve of early parameter values ​​and a second curve of current parameter values ​​for the N1 array of pairs, and determine the intersection point between the first curve and the second curve; If the number of intersection points is 0 or 1, the current parameter value of the corresponding analysis parameter will be retained; If the number of intersection points is not 0 and not 1, the first curve and the second curve are vertically divided according to the intersection points to sequentially obtain the curve area difference between every two intersection points; According to the area differences of all involved curves, the current parameter values ​​of the corresponding analysis parameters are adjusted to obtain new values ​​and retain them; The retained values ​​of different analysis parameters under the corresponding electrocardiogram are used as the HFimpEF prediction results of the corresponding patients.

2. The HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology according to claim 1, characterized in that: The physiological data include: blood pressure and blood oxygen of the corresponding patient.

3. The HFimpEF prediction method based on electrocardiogram combined with artificial intelligence technology according to claim 1, characterized in that: According to the area differences of all involved curves, the current parameter values ​​of the corresponding analysis parameters are adjusted to obtain new values ​​and retain them; ； in, Indicates the corresponding adjusted new value; Indicates the corresponding current parameter value; , and m1 represents the number of intersection points, and m2 represents the number of segmented curve segments based on the existence of intersection points; represents the curve area difference of the j2-th segmented curve segment; represents the curve area difference of the j3rd segment, and ; It represents the number of all curve area differences greater than 0; Indicates the number of all curve area differences that are less than 0.

Images