Multi-dimensional Auxiliary Screening Method for High-risk Populations of Coronary Heart Disease Combining Ultrasound Imaging

Abstract

The present invention relates to the technical field of coronary heart disease screening, and particularly to a multi-dimensional auxiliary screening method for high-risk populations of coronary heart disease combined with ultrasonic imaging. This method splits the components of the patient's heart rate data sequence, analyzes the proximity of the heart rate data on the heart rate segment, and determines the main component of the heart rate segment; and based on the abnormal condition of the beating frequency of the main component, obtains the abnormal probability to screen high-risk patients; determines the high-frequency video segment of the echocardiogram video based on the beating frequency of the high-risk patients, determines the diastolic direction vector through the continuous superposition of the left ventricular wall edges between echocardiograms in the high-frequency video segment, analyzes the chaotic change and decreasing trend in time series, and obtains the coronary heart disease risk probability. The present invention screens high-risk patients through multi-modal data fusion analysis of heart rate data and echocardiogram, and conducts subsequent analysis of abnormal left ventricular wall movement, improves the sensitivity of risk probability assessment, and enhances the accuracy and reliability of coronary heart disease risk assessment.

Description

Technical Field

[0001] The present invention relates to the technical field of coronary heart disease screening, and particularly to a multi-dimensional auxiliary screening method for high-risk populations of coronary heart disease combined with ultrasonic imaging. Background Art

[0002] With the rapid development of medical imaging, artificial intelligence, and big data technologies, relying solely on a single examination method can no longer meet the needs of modern medicine. Multi-dimensional joint analysis has become a new trend in the screening of high-risk populations of coronary heart disease. Comprehensive analysis by combining ultrasonic imaging with other biomarker and electrocardiogram data can provide a more comprehensive and accurate risk assessment. In particular, by introducing artificial intelligence technology and performing intelligent screening by combining different data sources, the accuracy of screening and the ability of early diagnosis can be effectively improved.

[0003] Existing multi-dimensional screenings often simply combine the detection results of different types of data to obtain screening results, without considering the fusion processing relationship of multi-modal data in different dimensions, and without fully exploiting the potential of multi-dimensional data, making the evaluation results of multi-dimensional data joint analysis unreliable. Summary of the Invention

[0004] In order to solve the technical problem of unreliable evaluation results in the joint analysis of multi-dimensional data in the prior art, the purpose of the present invention is to provide a multi-dimensional auxiliary screening method for high-risk populations of coronary heart disease combined with ultrasonic imaging, and the specific technical solution adopted is as follows:

[0005] The present invention provides a multi-dimensional auxiliary screening method for high-risk populations of coronary heart disease combined with ultrasonic imaging, and the method includes:

[0006] Obtaining a heart rate data sequence and an echocardiogram video of a sample patient during a detection period; dividing each heart rate data sequence into at least two heart rate segments; obtaining the IMF components of each heart rate data sequence;

[0007] For each sample patient, taking the maximum amplitude frequency of each IMF component in the frequency domain space as the main beating frequency of each IMF component; determining the main component of each heart rate segment according to the proximity between the heart rate data sequence and each IMF component in each heart rate segment; obtaining the abnormal probability of each sample patient according to the chaotic distribution of the main beating frequencies of the main components corresponding to all heart rate segments and the change fluctuation of the difference between the main beating frequencies of adjacent main components; determining high-risk patients from the sample patients based on the abnormal probability.

[0008] For each high-risk patient, the echocardiogram video is divided according to the magnitude distribution of the main beating frequency to determine the high-frequency video segment; according to the continuous superposition change of the left ventricular wall edge between adjacent echocardiograms in each high-frequency video segment, the left ventricular diastolic direction vector in each high-frequency video segment is determined; through the change chaos and the degree of decreasing trend of the left ventricular diastolic direction vector in time series, the coronary heart disease risk probability of the high-risk patient is obtained.

[0009] Further, the method for obtaining the main component includes:

[0010] For any heart rate segment, on the corresponding time period of this heart rate segment, the numerical differences between the heart rate data sequence and the elements of each IMF component at the same moment are arranged in chronological order to obtain the difference sequence between this heart rate segment and each IMF component;

[0011] According to the stability degree of the numerical distribution and the cumulative numerical size of the elements in the difference sequence between this heart rate segment and each IMF component, the closeness between this heart rate segment and each IMF component is obtained;

[0012] When the closeness between this heart rate segment and the IMF component is the largest, the corresponding IMF component is used as the main component of this heart rate segment.

[0013] Further, the method for obtaining the closeness includes:

[0014] Each IMF component is sequentially used as the target component; the numerical variance of all elements in the difference sequence between this heart rate segment and the target component is used as the deviation chaos degree between this heart rate segment and the target component;

[0015] The cumulative numerical value of all elements in the difference sequence between this heart rate segment and the target component is used as the deviation cumulative degree between this heart rate segment and the target component;

[0016] The product of the deviation chaos degree and the deviation cumulative degree between this heart rate segment and the target component is subjected to negative correlation mapping to obtain the closeness between this heart rate segment and the target component.

[0017] Further, the method for obtaining the abnormal probability includes:

[0018] For any sample patient, the main beating frequencies of the main components corresponding to all heart rate segments on the heart rate data sequence of this sample patient are arranged in chronological order to obtain the beating frequency sequence of this sample patient;

[0019] The numerical variance of all elements in the beating frequency sequence of this sample patient is used as the frequency chaos degree of this sample patient;

[0020] Calculate the mean difference between each element and its adjacent element in the beating frequency sequence of the sample patient to obtain the adjacent difference of each element; arrange the adjacent differences in the order of the elements in the beating frequency sequence to obtain the adjacent deviation sequence of the sample patient; for any element in the adjacent deviation sequence, calculate the numerical difference between this element and each other element in the adjacent deviation sequence, and then sum all the numerical differences to obtain the relative fluctuation degree of this element; when the relative fluctuation of all elements in the adjacent deviation sequence is the smallest, use the value of the corresponding element as the change fluctuation degree of the sample patient.

[0021] Multiply the frequency chaos degree and the change fluctuation degree of the sample patient as the abnormal probability of the sample patient.

[0022] Further, determining high-risk patients from sample patients based on the abnormal probability includes:

[0023] Take the sample patients with an abnormal probability greater than the preset abnormal threshold as high-risk patients.

[0024] Further, the method for obtaining the high-frequency video segment includes:

[0025] For any high-risk patient, arrange the elements in the beating frequency sequence of this high-risk patient in ascending order of numerical value to obtain a frequency ascending sequence.

[0026] Use the otsu threshold segmentation method to obtain the segmentation point of the frequency ascending sequence; take the main beating frequency greater than the segmentation point in the frequency ascending sequence as the high-frequency frequency.

[0027] Take the heart rate segment with the high-frequency frequency as the main component as the high-frequency time period, and divide the echocardiogram video in the high-frequency time period into high-frequency video segments.

[0028] Further, the method for obtaining the left ventricular diastolic direction vector includes:

[0029] Take the lower left corner of each echocardiogram as the origin and the sides of the echocardiogram as the coordinate axes to establish a two-dimensional image coordinate system; obtain the left ventricular wall edge in each echocardiogram.

[0030] For any high-frequency video segment, take the left ventricular wall edge in the first echocardiogram of this high-frequency video segment as the edge connected domain of the first echocardiogram.

[0031] For any echocardiogram in the high-frequency video segment except the first echocardiogram, when the abscissa of the center point of the left ventricular wall edge of this echocardiogram is greater than the abscissa of the center point of the left ventricular wall edge of the previous echocardiogram in time sequence, superimpose the left ventricular wall edge of this echocardiogram with the edge connectivity domain of the previous echocardiogram as the edge connectivity domain of this echocardiogram; otherwise, do not superimpose, and use the left ventricular wall edge of this echocardiogram as the new edge connectivity domain;

[0032] Traverse the echocardiograms in the high-frequency video segment in time sequence to obtain several edge connectivity domains; regard the edge connectivity domain that contains not only a single left ventricular wall edge as the diastolic connectivity domain;

[0033] Take the coordinates of all points of each diastolic connectivity domain in the image coordinate system as the input of the PCA algorithm to obtain the principal component vector as the diastolic direction vector of each left ventricle.

[0034] Further, the method for obtaining the coronary heart disease risk probability includes:

[0035] Take the angle when rotating clockwise along the horizontal upward direction to each left ventricular diastolic direction vector as the direction value of each left ventricular diastolic direction vector; take the vector value of each left ventricular diastolic direction vector as the motion value of each left ventricular diastolic direction vector;

[0036] Arrange the direction values of all left ventricular diastolic direction vectors in time sequence to obtain the diastolic direction sequence of high-risk patients; take the numerical variance of all elements in the diastolic direction sequence as the direction chaos degree of high-risk patients;

[0037] Map the motion values of all left ventricular diastolic direction vectors into the motion coordinate system, where the horizontal axis represents the time sequence and the vertical axis represents the motion value;

[0038] Take the coordinates of all data points in the motion coordinate system as the input of the PCA algorithm to obtain the principal component vector as the trend direction; perform negative correlation mapping and normalization processing on the angle when rotating counterclockwise along the vertical downward direction to the trend direction to obtain the trend attenuation degree of high-risk patients;

[0039] Combine the direction chaos degree and trend attenuation degree of high-risk patients to obtain the coronary heart disease risk probability of high-risk patients.

[0040] Further, the step of dividing each heart rate data sequence into at least two heart rate segments includes:

[0041] Use the otsu multi-threshold segmentation method for each heart rate data sequence to obtain several segmentation points for segmentation and obtain several heart rate segments.

[0042] Further, after obtaining the coronary heart disease risk probability of high-risk patients, it also includes:

[0043] Taking the data collected from high-risk patients as samples and the coronary heart disease risk probability as labels, an ensemble classifier is obtained through AdaBoost training; high-risk screening for coronary heart disease is performed through the trained ensemble classifier.

[0044] The present invention has the following beneficial effects:

[0045] The present invention improves the accuracy of coronary heart disease risk assessment by combining heart rate data in time series and echocardiograms, and considers the abnormal differences in the movement of the left ventricular wall between coronary heart disease patients and other patients for risk assessment. Since the difference in the continuous decrease of the left ventricular ejection fraction between coronary heart disease patients and the rest of the patients is more obvious when the heart rate increases, first, the data fluctuations of the heart rate data sequence are split through IMF components, the main beating conditions of the heart at different heart rate segments are analyzed, the main components at different heart rate segments are determined, and according to the abnormal conditions of the beating frequencies corresponding to the main components, the abnormal probability is obtained to screen high-risk patients. First, the instability of the increased heart rate is analyzed through heart rate data, and patients with significant heart rate abnormalities are screened out during sample analysis, so that during subsequent risk probability assessment, the risk sensitivity of sample data assessment is increased. Therefore, further combining ultrasonic images, the abnormal assessment and analysis of the movement conditions of the video segments at high frequencies determined by the main beating frequencies of high-risk patients are carried out. Through the continuous superposition of the left ventricular wall edges between echocardiograms in the high-frequency video segments, the diastolic condition of the left ventricular wall is reflected, the diastolic direction vector is determined, and finally, through the abnormal or weakened conditions of local wall motion, that is, through the chaotic change and decreasing trend of the diastolic direction vector, the coronary heart disease risk of the patient is evaluated. The present invention conducts multi-modal data fusion analysis of heart rate data and echocardiograms, screens high-risk patients based on heart rate data, increases the sensitivity of assessment and analysis, and then probabilistically evaluates the degree of abnormal movement of the left ventricular wall in the echocardiograms of high-risk patients, improving the accuracy and reliability of coronary heart disease risk assessment. Description of the Drawings

[0046] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a flowchart of a multi-dimensional auxiliary screening method for high-risk coronary heart disease population combined with ultrasonic images provided by an embodiment of the present invention;

[0048] Figure 2 It is a schematic diagram of an echocardiogram provided by an embodiment of the present invention;

[0049] Figure 3 Schematic diagram of the direction value of the left ventricular diastolic direction vector provided by an embodiment of the present invention;

[0050] Figure 4 Echocardiogram schematic diagram of the edge of segmental wall motion abnormality provided by an embodiment of the present invention. Detailed implementation manners

[0051] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following combines the accompanying drawings and preferred embodiments to detail the specific implementation manners, structures, features and effects of a multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic images according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0053] The following specifically describes the specific scheme of a multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic images provided by the present invention with reference to the accompanying drawings.

[0054] Coronary atherosclerotic heart disease, abbreviated as coronary heart disease, is caused by atherosclerotic lesions in the coronary artery intima, which narrow or occlude the lumen, resulting in weakened, disappeared or even paradoxical movement of the ventricular wall supplied by the artery. The left coronary artery and the right coronary artery are the two major arteries supplying blood to the myocardium. The left coronary artery is further divided into the left anterior descending branch and the left circumflex branch, which supply the anterior wall of the left ventricle, the anterior 2 / 3 of the interventricular septum, the lateral wall and the posterior wall of the left ventricle respectively, and the right coronary artery supplies the right ventricle, the posterior 1 / 3 of the interventricular septum and the inferior wall of the left ventricle. When the lumen of the blood vessel is narrowed due to atherosclerosis, it can cause insufficient blood supply to the corresponding area of the myocardium. If the lumen is suddenly blocked, severe and persistent myocardial ischemia can occur, leading to myocardial infarction.

[0055] Please refer to Figure 1 , which shows a flowchart of a multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic images provided by an embodiment of the present invention. The method includes the following steps:

[0056] S1: Obtain the heart rate data sequence and the echocardiogram video of the sample patient during the detection period; divide each heart rate data sequence into at least two heart rate segments; obtain the IMF components of each heart rate data sequence.

[0057] Echocardiography has high analytical value in judging the location of myocardial ischemia and myocardial infarction, as well as showing the complications after myocardial infarction. Therefore, coronary heart disease risk assessment can be carried out by obtaining ultrasonic images. The main differences in the echocardiogram between coronary heart disease patients and other patients are as follows:

[0058] 1. Changes in left ventricular function: The left ventricular function of ordinary patients may only show a transient decrease when the heart rate is too fast, while the left ventricular ejection fraction (LVEF) of coronary heart disease patients may continue to decrease and is more obvious when the heart rate increases.

[0059] 2. Abnormal left ventricular wall motion: The heart rate of ordinary patients being too fast will not show significant abnormal left ventricular wall motion, while coronary heart disease patients may have weakened or abnormal local ventricular wall motion due to myocardial ischemia or previous myocardial infarction.

[0060] Segmental ventricular wall motion abnormality is the main manifestation of coronary heart disease in echocardiography. Generally, the existing 20-segment division method is used to divide the ventricular wall into basal segments, middle segments, and apical segments. During myocardial infarction, segmental motion of the corresponding ventricular wall disappears or is significantly weakened, the diastolic wall thickening rate disappears, the heart cavity expands, the ventricular wall bulges, and the myocardial thickness thins, while the normal myocardial part shows compensatory enhanced motion, diastolic thickening, and increased amplitude.

[0061] For coronary heart disease patients, the incidence of tachycardia and bradycardia is relatively high. Tachycardia is associated with a poor prognosis. A too fast heart rate may be accompanied by an increase in myocardial oxygen consumption, thus easily inducing the occurrence of myocardial ischemia. Since the distinguishing features of coronary heart disease patients and other patients in the echocardiogram are significant under tachycardia, determining the tachycardia heart rate segment first can improve the subsequent evaluation results.

[0062] Therefore, in the embodiments of the present invention, heart rate data and echocardiography are continuously collected for each patient during the detection period to obtain a heart rate data sequence and an echocardiogram video. Among them, the heart rate data sequence is collected by an optical heart rate sensor or an electrical heart rate sensor. For example, a smart watch measures the change in blood flow at the wrist through an optical sensor to obtain heart rate data, and the heart rate data is recorded in the form of a time series, including time stamps and corresponding heart rate data values, forming a heart rate data sequence during the detection period. The echocardiogram video uses an ultrasonic probe to scan the heart at different positions on the chest to obtain ultrasonic images showing the heart structure to form a dynamic video, obtaining an echocardiogram video.

[0063] To facilitate the analysis of the heart rate beating state, the heart rate data sequence of each patient is initially segmented. In the embodiment of the present invention, the otsu multi-threshold segmentation is adopted for each heart rate data sequence to obtain several segmentation points for segmentation, and several heart rate segments are obtained. Each heart rate segment represents a heart rate with a similar frequency and represents a specific physiological activity state. For example, there are multiple different resting periods (low-frequency segments), exercise periods (medium-frequency segments), and intense exercise periods (high-frequency segments) in the heart rate data sequence, etc. It should be noted that the otsu multi-threshold segmentation method is a well-known technical means for those skilled in the art and will not be elaborated here.

[0064] Since the heart beating of the patient may be relatively chaotic, multiple IMF components are obtained by EMD decomposition of the heart rate data sequence. Each IMF component represents the heart rate vibration condition at a certain frequency, including normal heart rate beating, slower heart rate beating, and faster heart rate beating, which is convenient for subsequent analysis of the specific beating conditions of the heart rate segments.

[0065] S2: For each sample patient, the maximum amplitude frequency of each IMF component in the frequency domain space is used as the main beating frequency of each IMF component; on each heart rate segment, according to the proximity between the heart rate data sequence and each IMF component, the main component of each heart rate segment is determined; according to the chaotic condition of the main beating frequency distribution of the main components corresponding to all heart rate segments, and the change fluctuation condition of the difference between the main beating frequencies of adjacent main components, the abnormal probability of each sample patient is obtained; high-risk patients are determined from the sample patients based on the abnormal probability.

[0066] When assessing the risk of coronary heart disease, the risk sensitivity of different manifestation degrees is different. In order to improve the accuracy of risk assessment and increase the sensitivity to people with the risk of coronary heart disease during screening and assessment, first, the sample patients are screened. By screening out the population with a higher abnormal probability, that is, the population with tachycardia, the significant analysis accuracy of specific features can be improved in the subsequent analysis of the coronary heart disease characteristics in the image, the analysis of patients with unclear features can be reduced, and the assessment noise in the analyzed patient samples can be reduced.

[0067] Since the heart rate change of the patient is usually relatively complex, considering the beating frequency conditions of each component after decomposition, the maximum amplitude frequency of each IMF component in the frequency domain space is used as the main beating frequency of each IMF component. In the embodiment of the present invention, each IMF component is transformed into the frequency domain space through Fourier transform, and the frequency corresponding to the maximum amplitude is recorded as the main beating frequency of each IMF component. Each IMF component characterizes the frequency characteristics of the signal at different time scales, and the main beating frequency is helpful for multi-scale analysis and comprehensively evaluating the characteristics of heart rate beating from different scale perspectives.

[0068] Furthermore, based on the approximation of the decomposition results in each heart rate segment, the beating condition of each heart rate segment is analyzed locally, which is convenient for effectively reflecting the stable fluctuation of the beating frequency in the overall detection time subsequently. Therefore, on each heart rate segment, according to the proximity between the heart rate data sequence and each IMF component, the main component of each heart rate segment is determined, including:

[0069] First, for any heart rate segment, in the corresponding time period of this heart rate segment, the numerical differences between the elements of the heart rate data sequence and each IMF component at the same moment are arranged in chronological order to obtain the difference sequence between this heart rate segment and each IMF component. In the heart rate segment, the heart rate data sequence and each IMF component have corresponding numerical values at each moment. Through subtraction calculation and arranging the differences in the chronological order of the corresponding subtraction data, a difference sequence can be formed to reflect the continuous deviation of the numerical distributions of the heart rate data sequence and each IMF component.

[0070] Furthermore, the fitting condition between the IMF component and the heart rate data is reflected by the numerical values of the elements in the difference sequence. When the numerical values of the elements in the difference sequence are more consistent and the overall numerical value is smaller, it indicates that the IMF component fits better with the heart rate data. One element in the sequence corresponds to one data.

[0071] Therefore, in the embodiment of the present invention, according to the stability degree of the numerical distribution and the cumulative numerical size of the elements in the difference sequence between this heart rate segment and each IMF component, the proximity degree between this heart rate segment and each IMF component is obtained, including:

[0072] Each IMF component is sequentially used as the target component, and the proximity degree analysis is performed on each component. First, the variance of the numerical values of all elements in the difference sequence between this heart rate segment and the target component is used as the deviation chaos degree between this heart rate segment and the target component. The stability degree of the numerical distribution of the elements is reflected by the variance. When the variance is larger, that is, the deviation chaos degree is higher, it indicates that the difference degree between the heart rate data and the target component on the heart rate segment is not stable, and the proximity degree is lower.

[0073] Furthermore, the cumulative value of the numerical values of all elements in the difference sequence between this heart rate segment and the target component is used as the deviation accumulation degree between this heart rate segment and the target component. By accumulating the differences in the difference sequence, the overall deviation size in the heart rate segment is reflected. When the deviation accumulation degree is larger, it indicates that the overall difference degree is high, and the deviation between the heart rate data and the target component on the heart rate segment is higher, and the proximity degree is lower.

[0074] Therefore, the product of the deviation chaos degree and the deviation accumulation degree between this heart rate segment and the target component is subjected to a negative correlation mapping to obtain the proximity degree between this heart rate segment and the target component. The fitting degree is reflected by the negative correlation of the two deviation indexes. When the deviation chaos degree and the deviation accumulation degree are smaller, it reflects that the deviation is more consistent and low, then the component is closer to the heart rate data, and the proximity degree is larger. As an example, the expression of the proximity degree is:

[0075] D n,m = exp(-U n,m ×V n,m ); where D n,m represents the closeness between the nth heart rate segment and the mth IMF component, U n,m represents the deviation chaos degree between the nth heart rate segment and the mth IMF component, V n,m represents the deviation accumulation degree between the nth heart rate segment and the mth IMF component, and exp() represents the exponential function with the natural constant as the base.

[0076] It should be noted that negative correlation mapping is a well-known technical means in the art, and can be in the form of inverse ratio or negative exponential power, etc., which will not be limited and elaborated here.

[0077] Through the closeness analyzed with all IMF components, the IMF component that best fits the heart rate data in this heart rate segment can be determined. Finally, when the closeness between this heart rate segment and the IMF component is the largest, the corresponding IMF component is used as the main component of this heart rate segment.

[0078] So far, the determination of the main component of the heart rate segment is completed, and the main beating frequency of each heart rate segment is characterized by the beating frequency of the main component. According to the beating conditions of multiple heart rate segments, the abnormal significant conditions of the patient are analyzed. When the beating frequency of the patient is too chaotic and fluctuates violently, the higher the beating frequency, the more significant the heart rate abnormality of the patient, and there may be a more significant manifestation of coronary heart disease symptoms. Therefore, further analyze the distribution chaos and fluctuation of the main beating frequencies of all heart rate segments for each sample patient to obtain the abnormal probability.

[0079] Preferably, in the embodiment of the present invention, the method for obtaining the abnormal probability includes:

[0080] First, for any sample patient, arrange the main beating frequencies of the main components corresponding to all heart rate segments in the heart rate data sequence of this sample patient in chronological order to obtain the beating frequency sequence of this sample patient, which reflects the continuous situation of the close beating frequencies in the whole time series. For a more normal patient, the data in the beating frequency sequence will change more consistently and stably.

[0081] Further, take the numerical variance of all elements in the beating frequency sequence of this sample patient as the frequency chaos degree of this sample patient, and reflect the numerical dispersion of the elements in the beating frequency sequence through the numerical variance. When the variance is larger, it indicates that the stability of the beating frequency sequence is worse, and the abnormal situation of the sample patient is more significant.

[0082] After analyzing the overall data fluctuation by variance analysis, the stability of frequency changes is further analyzed from local fluctuations. First, calculate the mean difference between each element and its adjacent elements in the heart rate sequence of the sample patients to obtain the adjacent difference of each element. Through the adjacent difference situation, the local change degree of a single element is reflected. Except for the first and last elements, there are two adjacent elements, left and right, for each element in the sequence. Therefore, the adjacent change degree is reflected by the mean value. Then, sort the adjacent differences, arrange the adjacent differences in the order of the elements in the heart rate sequence to obtain the adjacent deviation sequence of the sample patients, and analyze the stability of local changes.

[0083] Furthermore, for any element in the adjacent deviation sequence, calculate the numerical difference between this element and each other element in the adjacent deviation sequence, and then sum all the numerical differences to obtain the relative fluctuation degree of this element. Through the degree of difference between each element and other elements, the relative instability degree of each element is reflected. When the sum of the numerical differences is larger, that is, the relative fluctuation is larger, it means that the distribution deviation degree of this element relative to other elements is larger. As an example, the expression of the relative fluctuation degree is:

[0084] In the formula, S i represents the relative fluctuation degree of the i-th element in the adjacent deviation sequence, x i represents the adjacent difference of the i-th element in the adjacent deviation sequence, x z represents the adjacent difference of the z-th other element except the i-th element in the adjacent deviation sequence, N represents the total number of other elements except the i-th element in the adjacent deviation sequence, and || represents the absolute value extraction function.

[0085] Furthermore, when the relative fluctuations of all elements in the adjacent deviation sequence are the smallest, the corresponding element value is used as the change fluctuation degree of the sample patients. By selecting the element with the smallest relative fluctuation, it reflects that this element is less affected by fluctuations in the sequence. At this time, the corresponding value of the element, that is, the adjacent difference, can better reflect the local change degree in the sequence. When the value of this element is smaller, that is, the change fluctuation degree is smaller, it means that the relative change of the sequence is smoother and the normality may be higher.

[0086] Finally, the product of the frequency chaos degree and the change fluctuation degree of the sample patients is used as the abnormal probability of the sample patients. Through comprehensive analysis of the overall and local parts in the sequence, when the frequency chaos degree and the change fluctuation degree are larger, it means that the continuous heart rate beating situation in time series is more unstable, the abnormal probability is higher, and the abnormal reflection of the sample patients' heart rate is more significant, and the abnormal performance in subsequent coronary heart disease may be more significant.

[0087] Furthermore, based on the abnormal probability, a preliminary screening is carried out. In the embodiments of the present invention, the sample patients with an abnormal probability greater than the preset abnormal threshold are regarded as high-risk patients. In the embodiments of the present invention, after normalizing the abnormal probability, an abnormal threshold of 0.7 is set for threshold judgment to screen the high-risk patients for further analysis. It should be noted that normalization is a technical means well-known to those skilled in the art, and the choice of normalization can be linear normalization or standard normalization, etc. The specific normalization method is not limited herein.

[0088] S3: For each high-risk patient, divide the echocardiogram video according to the magnitude distribution of the main beating frequency to determine the high-frequency video segment; determine the left ventricular diastolic direction vector in each high-frequency video segment according to the continuous superposition change of the left ventricular wall edge between adjacent echocardiograms; obtain the coronary heart disease risk probability of the high-risk patient through the chaotic situation of the change of the left ventricular diastolic direction vector in time series and the degree of decreasing trend.

[0089] By screening out the high-risk patients, the patients with abnormal heart rate beats in the sample patients are screened out. The characteristics of abnormal movement of the left ventricular wall in the echocardiogram of the high-risk patients with coronary heart disease are more significant. At this time, the sensitivity of the coronary heart disease risk probability evaluated based on the high-risk patients is better, and the probability analysis is more accurate.

[0090] First, analyze the time period with a higher frequency of heart rate beats screened out for the high-risk patients. The analysis of ventricular abnormal movement performance in this high-frequency time period is more reliable. Therefore, divide the echocardiogram video according to the magnitude distribution of the main beating frequency to determine the high-frequency video segment, including:

[0091] For any high-risk patient, arrange the elements in the beating frequency sequence of the high-risk patient in ascending order of numerical value to obtain an ascending frequency sequence. Then, use the otsu threshold segmentation method to obtain the segmentation point of the ascending frequency sequence, which reflects the position with the largest change between frequencies. Further, regard the main beating frequency greater than the segmentation point in the ascending frequency sequence as the high-frequency frequency, and the time period corresponding to the high-frequency frequency is the time period for subsequent main abnormal characterization analysis.

[0092] Therefore, regard the heart rate segment with the high-frequency frequency as the main component as the high-frequency time period, and divide the echocardiogram video in the high-frequency time period into high-frequency video segments. Each high-frequency video segment contains multiple frames of echocardiograms. Please refer to Figure 2 which shows a schematic diagram of an echocardiogram provided by an embodiment of the present invention.

[0093] As the respiration changes, when the manifestation of coronary heart disease is significant, the change amplitude at the edge of the left ventricular wall of the heart will show a significant weakening and deformation, while when the manifestation of coronary heart disease is weak, the change at the edge of the left ventricular wall shows a more regular change. Multiple left ventricular diastolic processes will occur in each high-frequency video segment. When there is a manifestation of coronary heart disease, the systolic thickening rate of the left ventricular wall of the heart disappears, the heart cavity expands, the ventricular wall bulges, the myocardial thickness thins, and the diastolic change in the edge of the ventricular wall will be more chaotic in the reverse direction and show an inflation manifestation. Therefore, in the high-frequency video segment, according to the continuous change of the left ventricular wall edge between consecutive echocardiogram images, the diastolic condition under each diastolic condition can be determined, which is convenient for subsequent analysis of abnormal conditions.

[0094] Preferably, in the embodiment of the present invention, the method for obtaining the left ventricular diastolic direction vector includes:

[0095] First, taking the lower left corner of each echocardiogram as the origin and the sides of the echocardiogram as the coordinate axes, a two-dimensional image coordinate system is established. Establishing the coordinate system uniformly facilitates the representation of directions. Furthermore, the left ventricular wall edge in each echocardiogram is obtained for analyzing abnormal movements. In the embodiment of the present invention, the left ventricular wall edge in each echocardiogram is obtained through the EchoNet-Dynamic 2020 model. It should be noted that the EchoNet-Dynamic 2020 model is a computer vision model based on deep learning, aiming to automatically analyze and extract key structures and dynamic information in echocardiograms, especially the extraction of the left ventricular wall (LV wall) edge. The edge segmentation obtained by the EchoNet-Dynamic 2020 model is a well-known public technical means in the art and will not be elaborated here. Please refer to Figure 4 , which shows a schematic echocardiogram of segmental wall motion abnormality edge provided by the embodiment of the present invention.

[0096] Furthermore, for any high-frequency video segment, the left ventricular wall edge in the first echocardiogram of the high-frequency video segment is used as the edge connected domain of the first echocardiogram, and the subsequent cardiac diastolic process is analyzed starting from the first echocardiogram.

[0097] For any echocardiogram other than the first echocardiogram in the high-frequency video segment, when the abscissa of the center point of the left ventricular wall edge of this echocardiogram is greater than the abscissa of the center point of the left ventricular wall edge of the previous echocardiogram in time sequence, it indicates that the left ventricular wall has undergone an expansion change, and edge superposition can be performed to reflect the diastolic range. The left ventricular wall edge of this echocardiogram is superposed with the edge connected domain of the previous echocardiogram as the edge connected domain of this echocardiogram. Otherwise, it is not an expansion change, no superposition is performed, the connected domain obtained previously stops, and the left ventricular wall edge of this echocardiogram is used as a new edge connected domain for subsequent superposition analysis.

[0098] Traverse the echocardiograms in this high-frequency video segment in chronological order to obtain several edge-connected regions, and analyze the superposition situation for each echocardiogram. Furthermore, the edge-connected regions that contain not only the edges of a single left ventricular wall are used as diastolic-connected regions, and the wall contraction situation with only a single left ventricular wall edge is usually discontinuous diastolic change. Each diastolic-connected region represents a left ventricular diastolic process.

[0099] Finally, take the coordinates of all points in the image coordinate system of each diastolic-connected region as the input of the PCA algorithm, and obtain the principal component vector as the left ventricular diastolic direction vector for each. It should be noted that PCA is a commonly used dimensionality reduction and feature extraction technology that extracts the vectors of the main changes in the data. Taking the coordinates of all points as the input for PCA analysis, that is, performing dimensionality reduction and direction analysis on the distribution of all data points. The obtained principal component vector represents the main direction and degree of the morphological changes of this connected region, reflecting the most significant shape change situation of the left ventricle during diastole. The PCA algorithm is a well-known technical means familiar to those skilled in the art and will not be elaborated here.

[0100] Multiple left ventricular diastolic direction vectors are obtained through multiple diastolic-connected regions, reflecting the main diastolic conditions of each left ventricular diastole. Since the cardiac cavity expands significantly at a higher heart rate in patients with coronary heart disease, the ventricular wall will bulge and cause deformation of the ventricle. Therefore, the diastolic directions of different diastolic processes will be more chaotic. And due to myocardial abnormalities, the segmental movement of the ventricular wall will be significantly weakened or disappear. Therefore, the movement trend in the diastolic direction will show a decreasing state.

[0101] Preferably, in the embodiments of the present invention, the coronary heart disease risk probability of high-risk patients is obtained through the chaotic situation of the changes in the left ventricular diastolic direction vectors in time series and the degree of decreasing trend of the left ventricular diastolic direction vectors, including:

[0102] First, the angle when rotating clockwise along the upward horizontal direction to each left ventricular diastolic direction vector is used as the direction value of each left ventricular diastolic direction vector, and the angle is assigned to each left ventricular diastolic direction by fixing the upward horizontal direction. The vector value of each left ventricular diastolic direction vector is used as the movement value of each left ventricular diastolic direction vector, and the segmental movement degree of the ventricular wall is reflected by the vector magnitude. Please refer to Figure 3 which shows a schematic diagram of the direction value of a left ventricular diastolic direction vector provided by an embodiment of the present invention.

[0103] Arrange the direction values of all left ventricular diastolic direction vectors in chronological order to obtain the diastolic direction sequence of high-risk patients, which represents the main diastolic direction changes of the ventricular wall in time series. The numerical variance of all elements in the diastolic direction sequence is used as the direction chaos degree of high-risk patients. When the variance is larger, that is, the direction chaos degree is larger, it indicates that there are more changes in the diastolic direction in time series and the coronary heart disease abnormality is more significant.

[0104] Map the motion values of all left ventricular diastolic direction vectors to the motion coordinate system. The horizontal axis represents the sequence value in the time series, and the vertical axis represents the motion value. Map the motion conditions in the order of time distribution in the motion coordinate system to reflect the change of the motion value.

[0105] Furthermore, take the coordinates of all data points in the motion coordinate system as the input of the PCA algorithm to obtain the principal component vector as the trend direction. Reduce the influence of local complex changes on the overall trend analysis through principal component analysis to obtain the change direction of the motion value. Further perform negative correlation mapping and normalization processing on the angle when rotating counterclockwise to the trend direction along the vertically downward direction to obtain the trend attenuation degree of high-risk patients. When the trend direction is closer to the vertically downward direction, that is, the smaller the angle, the more obvious and significant the weakening degree, and the more significant the coronary heart disease manifestation.

[0106] Finally, combine the direction chaos degree and trend attenuation degree of high-risk patients to obtain the coronary heart disease risk probability of high-risk patients. In the embodiment of the present invention, the product of the direction chaos degree and trend attenuation degree is used as the coronary heart disease risk probability of high-risk patients. When the direction chaos degree and trend attenuation degree indicate that the abnormal movement of coronary heart disease reflected by the echocardiogram video is more significant, the higher the coronary heart disease probability. As an example, the expression of the coronary heart disease risk probability is:

[0107] P k = F k ×(1 - norm(G k )); In the formula, P k represents the coronary heart disease risk probability of the k-th high-risk patient, F k represents the direction chaos degree of the k-th high-risk patient, G k represents the angle between rotating counterclockwise to the trend direction of the k-th high-risk patient along the vertically downward direction, norm() represents the normalization function, and (1 - norm(G k )) represents the trend attenuation degree of the k-th high-risk patient.

[0108] So far, through the fusion analysis of multi-modal heart rate data and echocardiogram data, a more sensitive coronary heart disease risk probability is obtained, improving the accurate results of multi-dimensional evaluation.

[0109] Furthermore, the sample data can be further used for model training, and the trained model is used for screening the coronary heart disease risk of the population to improve the accuracy and reliability of the screening. In the embodiment of the present invention, after obtaining the coronary heart disease risk probability of high-risk patients, the data collected from high-risk patients is used as a sample. The collected data must include heart rate data and echocardiogram. In other embodiments of the present invention, other data such as magnetic resonance imaging or blood pressure can also be combined as inputs. Take the coronary heart disease risk probability as a label and obtain an ensemble classifier through AdaBoost training.

[0110] AdaBoost (Adaptive Boosting) is an ensemble learning method that obtains a strong classifier by combining multiple weak classifiers. A weak classifier refers to a classifier whose performance is slightly better than random guessing, while an ensemble classifier obtains a stronger classification ability by weighted combination of the results of multiple weak classifiers. During training, each weak classifier makes predictions and gives results based on different features of the input data, such as heart rate data and echocardiogram. According to the weighted output of the weak classifiers, the final prediction result is obtained. It should be noted that training an ensemble classifier by AdaBoost is a well-known technical means for those skilled in the art and will not be elaborated here.

[0111] For coronary heart disease high-risk screening using the trained ensemble classifier, when there is new patient data, including heart rate data and echocardiogram, or other data, is input into the ensemble classifier, a predicted probability of coronary heart disease will be obtained through the classifier, reflecting the possible risk of coronary heart disease.

[0112] In summary, the present invention improves the accuracy of coronary heart disease risk assessment by combining heart rate data and echocardiogram in time series, and considers the abnormal differences in the left ventricular wall movement between coronary heart disease patients and other patients for risk assessment. Since the difference in the continuous decrease of left ventricular ejection fraction between coronary heart disease patients and the rest of the patients is more obvious when the heart rate increases, first, the data fluctuations of the heart rate data sequence are split by IMF components, the main beating conditions of the heart at different heart rate segments are analyzed, the main components at different heart rate segments are determined, and according to the abnormal conditions of the beating frequencies corresponding to the main components, the abnormal probability is obtained to screen high-risk patients. First, the instability of the increased heart rate is analyzed through heart rate data, and patients with significantly abnormal heart rates are screened out during sample analysis, so as to increase the risk sensitivity of sample data evaluation during subsequent risk probability assessment. Therefore, further combined with ultrasonic images, the abnormal evaluation and analysis of the movement conditions of the video segments at high frequencies determined by the main beating frequencies of high-risk patients are carried out. Through the continuous superposition of the left ventricular wall edges between echocardiograms in the high-frequency video segments, the diastolic condition of the left ventricular wall is reflected, the diastolic direction vector is determined, and finally, the risk of coronary heart disease in patients is evaluated through the abnormal or weakened conditions of local wall motion, that is, through the chaotic change and decreasing trend of the diastolic direction vector. The present invention analyzes the multi-modal data fusion of heart rate data and echocardiogram, screens high-risk patients based on heart rate data, increases the sensitivity of evaluation and analysis, and then probabilistically evaluates the abnormal degree of left ventricular wall movement in the echocardiogram of high-risk patients, improving the accuracy and reliability of coronary heart disease risk assessment.

[0113] It should be noted that: The above order of the embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the particular order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0114] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. The key point of each embodiment is to illustrate the differences from other embodiments.

Claims

1. A multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasound imaging, characterized in that, The method includes: Obtaining the heart rate data sequence and echocardiogram video of a sample patient during a detection period; dividing each heart rate data sequence into at least two heart rate segments; obtaining the IMF components of each heart rate data sequence; For each sample patient, taking the maximum amplitude frequency of each IMF component in the frequency domain space as the main beating frequency of each IMF component; determining the main component of each heart rate segment according to the proximity between the heart rate data sequence and each IMF component on each heart rate segment; for any sample patient, arranging the main beating frequencies of the main components corresponding to all heart rate segments on the heart rate data sequence of this sample patient in chronological order to obtain the beating frequency sequence of this sample patient; taking the numerical variance of all elements in the beating frequency sequence of this sample patient as the frequency chaos degree of this sample patient; calculating the mean difference between each element and its adjacent element in the beating frequency sequence of this sample patient to obtain the adjacent difference of each element; arranging the adjacent differences in the order of element distribution in the beating frequency sequence to obtain the adjacent deviation sequence of this sample patient; for any element in the adjacent deviation sequence, calculating the numerical difference between this element and each other element in the adjacent deviation sequence and then summing all the numerical differences to obtain the relative fluctuation degree of this element; when the relative fluctuation of all elements in the adjacent deviation sequence is the smallest, taking the value of the corresponding element as the change fluctuation degree of this sample patient; taking the product of the frequency chaos degree and the change fluctuation degree of this sample patient as the abnormal probability of this sample patient; determining high-risk patients from the sample patients based on the abnormal probability; For each high-risk patient, dividing the echocardiogram video according to the magnitude distribution of the main beating frequencies to determine the high-frequency video segments; determining the left ventricular diastolic direction vector in each high-frequency video segment according to the continuous superposition change of the left ventricular wall edge between adjacent echocardiograms in each high-frequency video segment; Taking the angle when rotating clockwise along the horizontal upward direction to each left ventricular diastolic direction vector as the direction value of each left ventricular diastolic direction vector; taking the vector value of each left ventricular diastolic direction vector as the motion value of each left ventricular diastolic direction vector; arranging the direction values of all left ventricular diastolic direction vectors in chronological order to obtain the diastolic direction sequence of the high-risk patient; taking the numerical variance of all elements in the diastolic direction sequence as the direction chaos degree of the high-risk patient; mapping the motion values of all left ventricular diastolic direction vectors to the motion coordinate system, with the horizontal axis representing the time sequence and the vertical axis representing the motion value; taking the coordinates of all data points in the motion coordinate system as the input of the PCA algorithm to obtain the principal component vector as the trend direction; performing negative correlation mapping and normalization processing on the angle when rotating counterclockwise along the vertical downward direction to the trend direction to obtain the trend attenuation degree of the high-risk patient; combining the direction chaos degree and the trend attenuation degree of the high-risk patient to obtain the coronary heart disease risk probability of the high-risk patient.

2. The multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic imaging according to claim 1, characterized in that, The method for obtaining the main component includes: For any heart rate segment, on the corresponding time period of the heart rate segment, arrange the numerical differences between the elements of the heart rate data sequence and each IMF component at the same moment in chronological order to obtain the difference sequence between the heart rate segment and each IMF component; Obtain the closeness degree between the heart rate segment and each IMF component according to the numerical distribution stability and numerical accumulation size of the elements in the difference sequence between the heart rate segment and each IMF component; When the closeness degree between the heart rate segment and the IMF component is the largest, take the corresponding IMF component as the main component of the heart rate segment.

3. The multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic imaging according to claim 2, wherein The method for obtaining the closeness degree includes: Successively take each IMF component as the target component; take the numerical variance of all elements in the difference sequence between the heart rate segment and the target component as the deviation chaos degree between the heart rate segment and the target component; Take the numerical accumulation value of all elements in the difference sequence between the heart rate segment and the target component as the deviation accumulation degree between the heart rate segment and the target component; Perform a negative correlation mapping on the product of the deviation chaos degree and the deviation accumulation degree between the heart rate segment and the target component to obtain the closeness degree between the heart rate segment and the target component.

4. The multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic imaging according to claim 1, wherein Determining high-risk patients from sample patients based on the abnormal probability includes: Take the sample patients with abnormal probability greater than the preset abnormal threshold as high-risk patients.

5. The multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic imaging according to claim 1, characterized in that, The method for obtaining the high-frequency video segment includes: For any high-risk patient, arrange the elements in the beating frequency sequence of the high-risk patient in ascending order of numerical value to obtain an ascending frequency sequence; Use the otsu threshold segmentation method to obtain the segmentation point of the ascending frequency sequence; take the main beating frequency greater than the segmentation point in the ascending frequency sequence as the high-frequency; Take the heart rate segment with the high-frequency as the main component as the high-frequency time period, and divide the echocardiogram video in the high-frequency time period into high-frequency video segments.

6. The multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic imaging according to claim 1, wherein The method for obtaining the left ventricular diastolic direction vector includes: Taking the lower left corner of each echocardiogram as the origin and the sides of the echocardiogram as the coordinate axes, establish a two-dimensional image coordinate system; obtain the left ventricular wall edge in each echocardiogram; For any high-frequency video segment, take the left ventricular wall edge in the first echocardiogram of the high-frequency video segment as the edge connected region of the first echocardiogram; For any echocardiogram other than the first echocardiogram in the high-frequency video segment, when the abscissa of the center point of the left ventricular wall edge of the echocardiogram is greater than the abscissa of the center point of the left ventricular wall edge of the previous echocardiogram in time sequence, superimpose the left ventricular wall edge of the echocardiogram on the edge connected region of the previous echocardiogram as the edge connected region of the echocardiogram; otherwise, do not superimpose, and take the left ventricular wall edge of the echocardiogram as the new edge connected region; Traverse the echocardiograms in the high-frequency video segment in time sequence to obtain several edge connected regions; take the edge connected region that contains not only a single left ventricular wall edge as the diastolic connected region; Take the coordinates of all points in each diastolic connected region in the image coordinate system as the input of the PCA algorithm, and obtain the principal component vector as each left ventricular diastolic direction vector.

7. The multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic imaging according to claim 1, characterized in that, Dividing each heart rate data sequence into at least two heart rate segments includes: For each heart rate data sequence, the Otsu multi-threshold segmentation method is used to obtain several segmentation points for segmentation, and several heart rate segments are obtained.

8. The multi-dimensional auxiliary screening method for high-risk coronary heart disease patients combined with ultrasonic imaging according to claim 1, characterized in that, After obtaining the coronary heart disease risk probability of high-risk patients, the following steps are further included: Taking the data collected from high-risk patients as samples and the coronary heart disease risk probability as labels, an ensemble classifier is obtained through AdaBoost training; coronary heart disease high-risk screening is performed through the trained ensemble classifier.

Images