Heart sound abnormity identification method for early acute heart failure patient

Abstract

The invention relates to the technical field of heart sound health detection, in particular to a heart sound abnormality recognition method for an early acute heart failure patient. The method comprises the following steps: acquiring a first heart sound moment and a second heart sound moment of a sample heart sound signal, and determining a window combination based on position distribution of two sliding windows on the sample heart sound signal; according to the similarity between the cardiac cycle of a short sliding window in the window combination of the sample heart sound signal and the neighborhood cardiac cycle and the distance between the short sliding window and the cardiac window cycle of the short sliding window, fusing the feature vectors of the sliding windows of the window combination to obtain a fusion vector; and constructing a hidden Markov model based on the fusion vector of the window combination of the sample heart sound signals, segmenting the heart sound signals of the patient to be detected by using the hidden Markov model, and further carrying out heart sound anomaly detection on the patient to be detected. According to the method, the features of the sliding windows with different scales of the sample heart sound signals are fused, so that the segmentation accuracy of the heart sound signals is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of heart sound health detection, and particularly to a method for identifying abnormal heart sounds in patients with early acute heart failure. Background Art

[0002] Since acute heart failure has a sudden onset and a critical condition, it is crucial to determine it early to quickly initiate a targeted treatment plan for acute heart failure. In the case of acute heart failure, both the function and structure of the heart are affected, resulting in abnormal changes in heart sounds. In order to accurately analyze and identify the abnormal manifestations of heart sound signals, it is necessary to segment the heart sound signals.

[0003] When using the traditional hidden Markov model to segment heart sound signals, the features of heart sound signals are extracted by using a sliding window with a fixed length to generate an observation value sequence, so as to achieve the segmentation of heart sound signals. However, since the heart sound signals of heart failure patients contain various abnormal components and the overall heart sound signals are more complex, it is difficult for the sliding window with a fixed length to take into account the features of different components of heart sound signals. A shorter sliding window may not contain enough information to judge the overall state of heart sounds, and a longer sliding window is likely to lose the abnormally occurring heart sounds that appear briefly or the features that change rapidly, thus making the division of heart sound segments of heart sound signals inaccurate and reducing the accuracy of identifying abnormal heart sounds in patients with acute heart failure. Summary of the Invention

[0004] In order to solve the technical problem that the sliding window with a fixed length in the hidden Markov model is difficult to take into account the features of different components of heart sound signals, resulting in inaccurate division of heart sound segments of heart sound signals, the purpose of the present invention is to provide a method for identifying abnormal heart sounds in patients with early acute heart failure, and the specific technical solution adopted is as follows: The present invention proposes a method for identifying abnormal heart sounds in patients with early acute heart failure, and the method includes: Obtain the heart sound signal of the patient to be tested and the sample heart sound signal, and the sample heart sound signal is divided into four heart sound segments; Use different types of sliding windows to traverse each sample heart sound signal, and extract the feature vectors of each sliding window on the sample heart sound signal; according to the amplitude distribution and duration of different heart sound segments of each sample heart sound signal, obtain the first heart sound moment and the second heart sound moment of each sample heart sound signal; The time period between two adjacent first heart sound moments of each sample heart sound signal is recorded as a cardiac cycle; window combinations are determined based on the position distributions of two types of sliding windows on each sample heart sound signal; according to the amplitude similarity between the cardiac cycle in which the short sliding window is located and its neighboring cardiac cycles in each window combination of each sample heart sound signal, and the distance between the short sliding window and the cardiac cycle in which it is located, the feature vectors of the two sliding windows in each window combination are fused to obtain a fusion vector for each window combination of each sample heart sound signal; An implicit Markov model is constructed based on the fusion vectors of the window combinations of the sample heart sound signals, and the heart sound signal of the patient to be measured is divided into different types of heart sound segments by using the implicit Markov model; heart sound abnormality detection is performed on the patient to be measured based on the heart sound segments of the heart sound signal.

[0005] Further, the obtaining of the first heart sound moment and the second heart sound moment of each sample heart sound signal includes: The types of heart sound segments include a first heart sound segment, a second heart sound segment, a third heart sound segment, and a fourth heart sound segment; A target sliding window is set, and the length of the target sliding window is the minimum value among the durations of the first heart sound segment and the second heart sound segment of all sample heart sound signals; each sample heart sound signal is traversed by using the target sliding window, and the peaks of the signals within all the target sliding windows on each sample heart sound signal are arranged in time sequence to obtain an amplitude sequence; For the remaining elements in the amplitude sequence except for a continuous number of elements smaller than the judgment threshold, if each element is greater than its adjacent previous element, then each element is recorded as a target element; the judgment threshold is equal to the amplitude mean value of all sample points of the third heart sound segment and the fourth heart sound segment of all sample heart sound signals; According to the amplitude difference of the data points within the preset neighborhood range of the data points corresponding to the adjacent target elements in the amplitude sequence of each sample heart sound signal, the moments corresponding to the target elements in the amplitude sequence are divided into a first heart sound time period and a second heart sound moment.

[0006] Further, the dividing of the moments corresponding to the target elements in the amplitude sequence into a first heart sound time period and a second heart sound moment includes: The target elements in the amplitude sequence of each sample heart sound signal are arranged in order to obtain a target sequence; The amplitude mean value of all data points within the preset neighborhood range of the data points corresponding to each element in the target sequence in each sample heart sound signal is calculated as the neighborhood amplitude of the corresponding element; Divide the target sequence into subsequences of length 2. If the neighborhood amplitude of the first element in the subsequence is greater than that of the second element, mark the subsequence as a target subsequence. Determine whether the number of target subsequences in each sample heart sound signal is greater than the number of non-target subsequences. If so, mark the time corresponding to the first element in the subsequences of the target sequence as the first heart sound time; if not, mark the time corresponding to the second element in the subsequences of the target sequence as the second heart sound time.

[0007] Further, the method for obtaining the fusion vector of each window combination of each sample heart sound signal includes: Obtain the distance sensitivity of the sliding window within the corresponding cardiac cycle according to the amplitude similarity between each cardiac cycle of each sample heart sound signal and its neighboring cardiac cycles. Denote the time interval between the middle moment within the short sliding window and the start moment within the cardiac cycle where the short sliding window is located in each window combination of each sample heart sound signal as the judgment distance corresponding to the window combination. Use the distance sensitivity of the cardiac cycle where the short sliding window is located in each window combination of each sample heart sound signal to adjust the judgment distance of each window combination, and obtain the fusion coefficient of the short sliding window in the corresponding window combination. Use the fusion coefficient to fuse the feature vectors of the two sliding windows in each window combination of each sample heart sound signal, and obtain the fusion vector of each window combination of each sample heart sound signal.

[0008] Further, the method for obtaining the distance sensitivity includes: Take the cumulative sum of the sum of the absolute values of the amplitude differences between each cardiac cycle of each sample heart sound signal and the start moment of its neighboring cardiac cycle and the absolute value of the amplitude difference at the second heart sound moment as the overall neighborhood difference degree of each cardiac cycle of each sample heart sound signal. Take the ratio of the neighborhood amplitudes of the amplitudes at the start moment of the cardiac cycle and the second heart sound moment as the neighborhood ratio. Denote the variance of the difference between each cardiac cycle of each sample heart sound signal and the neighborhood ratio of its neighboring cardiac cycles as the neighborhood discrete value of each cardiac cycle. Obtain the distance sensitivity of the sliding window within each cardiac cycle of each sample heart sound signal according to the overall neighborhood difference degree and the neighborhood discrete value.

[0009] Further, the method for obtaining the fusion coefficient includes: Use the distance sensitivity of the cardiac cycle where the short sliding window is located in each window combination of each sample heart sound signal to perform weighted processing on the judgment distance, perform negative correlation and normalization processing on the weighted result, and obtain the fusion coefficient of the short sliding window in the corresponding window combination.

[0010] Further, the fusion vector of each window combination of each sample heart sound signal includes: For each window combination of each sample heart sound signal, the fusion coefficient is respectively used as the weight of the feature vector of the short sliding window in the window combination, and the difference between the constant 1 and the fusion coefficient is used as the weight of the feature vector of the long sliding window in the window combination. Half of the result of weighted summation of the feature vectors of the short sliding window and the long sliding window in the window combination is used as the fusion vector of the window combination.

[0011] Further, constructing a hidden Markov model based on the fusion vector of the window combination of the sample heart sound signal includes: Sequentially arranging the fusion vectors of all window combinations of each sample heart sound signal to obtain the feature sequence of each sample heart sound signal; inputting the feature sequences of all sample heart sound signals into the hidden Markov model to construct the hidden Markov model.

[0012] Further, determining the window combination based on the position distribution of two types of sliding windows on each sample heart sound signal includes: For each sample heart sound signal, calculate the time interval between the middle moment of each short sliding window on the sample heart sound signal and all long sliding windows respectively. A window combination is formed by the long sliding window corresponding to the smallest such time interval and each short sliding window.

[0013] Further, the types of the sliding windows include: long sliding windows and short sliding windows.

[0014] The present invention has the following beneficial effects: In the embodiment of the present invention, the functional state of the heart in the case of heart failure will significantly affect the intensity and characteristics of heart sounds. The first heart sound and the second heart sound are more significantly affected by heart failure. The abnormal heart sound conditions can be accurately analyzed to determine the first heart sound moment and the second heart sound moment of the sample heart sound signal. In order to avoid that the fixed-length sliding window is difficult to take into account the characteristics of different components of the heart sound signal, multi-scale windows are used to process the sample heart sound signal. The amplitude similarity between the cardiac cycle and its neighboring cardiac cycles presents the sensitivity degree of the sliding window pair distance in the window combination of the cardiac cycle. Combining the distance between the short sliding window in the window combination and the cardiac cycle it is in, the feature vectors of the sliding windows of the window combination are adaptively fused, so that the fusion vector can better reflect the dynamic change process of the heart sound signal, which helps to more accurately identify the positions of various heart sounds in the cardiac cycle in the subsequent heart sound signal segmentation process. Thus, the accurate division of the heart sound segments of the heart sound signal is realized, and the accuracy of abnormal heart sound recognition for patients with acute heart failure is improved. Description of the Drawings

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0016] Figure 1 The flowchart of the steps of a method for identifying abnormal heart sounds in patients with early acute heart failure provided by an embodiment of the present invention; Figure 2 The flowchart of the steps of a method for obtaining a fusion vector provided by an embodiment of the present invention; Figure 3 The schematic diagram of a computer device of a device for identifying abnormal heart sounds in patients with early acute heart failure provided by an embodiment of the present invention. Detailed implementation manners

[0017] 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 method for identifying abnormal heart sounds in patients with early acute heart failure proposed 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.

[0018] 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.

[0019] The following specifically describes the specific solution of a method for identifying abnormal heart sounds in patients with early acute heart failure provided by the present invention with reference to the accompanying drawings.

[0020] Embodiment 1: The present invention proposes a method for identifying abnormal heart sounds in patients with early acute heart failure. Please refer to Figure 1 , which shows the flowchart of the steps of a method for identifying abnormal heart sounds in patients with early acute heart failure provided by an embodiment of the present invention. The method includes: Step S1: Obtain the heart sound signal of the patient to be tested and the sample heart sound signal, and the sample heart sound signal is divided into four heart sound segments.

[0021] Specifically, a doctor uses a phonocardiograph with a built-in Bluetooth system to collect the heart sound signals of heart failure patients and uploads the heart sound signals of heart failure patients to the cloud database. Select the heart sound signals uploaded within the past month from the cloud database. To reduce the computational amount, use Adobe Audition audio processing software to randomly intercept the signals of each heart sound signal for 1 continuous minute, which is recorded as the sample heart sound signal. At the same time, use the phonocardiograph to collect the heart sound signals of the patient to be tested for 20 minutes.

[0022] The heart sound signal of one cardiac cycle includes four heart sounds, which are, in the order of appearance, the first heart sound, the third heart sound, the second heart sound, and the fourth heart sound. Through the method of manual calibration, that is, the doctor segments and calibrates the sample heart sound signal, each sample heart sound signal is divided into four heart sound segments, including: the first heart sound segment, the second heart sound segment, the third heart sound segment, and the fourth heart sound segment. Among them, the first heart sound segment refers to the corresponding part of the first heart sound of the sample voice signal, and the meanings of other heart sound segments are similar to that of the first heart sound segment.

[0023] In a specific implementation manner of the embodiment of the present invention, the signal sampling frequency is set to 10,000 Hertz.

[0024] Step S2: Use different types of sliding windows to traverse each sample heart sound signal, and extract the feature vectors of each sliding window on the sample heart sound signal; according to the amplitude distribution and duration of different heart sound segments of each sample heart sound signal, obtain the first heart sound moment and the second heart sound moment of each sample heart sound signal.

[0025] In the process of segmenting the heart sound signal using the hidden Markov model, it is difficult to take into account the features of different components when extracting the features of the heart sound signal with a window of fixed length. In this solution, different types of sliding windows are used to traverse each sample heart sound signal, the feature vectors of the sliding windows are extracted, and the feature vectors extracted by the sliding windows of different scales are fused to construct a hidden Markov model, which improves the accuracy of the model for segmenting the heart sound signal.

[0026] In a specific implementation manner of the embodiment of the present invention, the types of sliding windows include: long sliding windows and short sliding windows, and the lengths of the long sliding windows and short sliding windows are set to 19 and 9 in sequence, and the implementer can set them according to the specific situation. It should be noted that all windows in this solution refer to time windows.

[0027] In an embodiment of the present invention, the method for obtaining the feature vector of the sliding window is as follows: Denote the signals within each sliding window of each type on the sample heart sound signal as window analysis signals, arrange the component signals obtained by decomposing the window analysis signals in the order of corresponding frequencies to obtain a frequency sequence; Calculate the sum of the squares of the amplitudes of the data points on the window analysis signal as the energy; Arrange the peak value, energy, and frequency sequence of the window analysis signal in order to obtain the feature vector of the sliding window corresponding to the window analysis signal.

[0028] In a specific implementation manner of the embodiment of the present invention, the wavelet transform algorithm is used to decompose the window analysis signal to obtain component signals of different frequencies.

[0029] In the case of heart failure, the functional state of the heart will significantly affect the intensity and characteristics of heart sounds. Since the amplitudes of the first heart sound and the second heart sound are relatively large and are more significantly affected by heart failure, the correlation degree between the first heart sound, the second heart sound, and heart failure is relatively large. In order to analyze the abnormal heart sound conditions of heart failure patients, it is necessary to determine the first heart sound moment and the second heart sound moment of the sample heart sound signal.

[0030] Preferably, in some possible implementation manners of the embodiment of the present invention, the method for obtaining the first heart sound period and the second heart sound moment includes: The types of heart sound segments include the first heart sound segment, the second heart sound, the third heart sound segment, and the fourth heart sound segment; Set a target sliding window, and the length of the target sliding window is the minimum value of the durations of the first heart sound segment and the second heart sound segment of all sample heart sound signals; Use the target sliding window to traverse each sample heart sound signal, arrange the peak values of the signals within all target sliding windows on each sample heart sound signal in time sequence to obtain an amplitude sequence; For the remaining elements in the amplitude sequence except for a continuous number of elements smaller than the judgment threshold, if each element is greater than its adjacent previous element, then each element is denoted as a target element; The judgment threshold is equal to the average amplitude of all sample points of the third heart sound segment and the fourth heart sound segment of all sample heart sound signals; According to the amplitude difference of the data points within the preset neighborhood range of the data points corresponding to the adjacent target elements in the amplitude sequence of each sample heart sound signal, divide the moments corresponding to the target elements in the amplitude sequence into the first heart sound period and the second heart sound moment.

[0031] To analyze the first heart sound segment and the second heart sound segment more precisely, it is necessary to ensure that the target sliding window contains the first heart sound segment or the second heart sound segment, but does not entirely contain the third heart sound segment or the fourth heart sound segment. Given that the durations of the first heart sound and the second heart sound are shorter than those of the third heart sound and the fourth heart sound, the length of the target sliding window is the minimum of the durations of the first heart sound segment and the second heart sound segment among all sample heart sound signals. The amplitudes of the first heart sound and the second heart sound are greater than those of the third heart sound and the fourth heart sound. Also, within one cardiac cycle, the order of the heart sounds is the first heart sound, the third heart sound, the second heart sound, and the fourth heart sound. If the signal peaks within a continuous preset number of target sliding windows are all less than the judgment threshold, it indicates that these target sliding windows may be located in the third heart sound segment or the fourth heart sound segment. Among the remaining target sliding windows, if the signal peak within a target sliding window is greater than the signal peak within its previous target sliding window, then this target sliding window is in the first heart sound segment or the second heart sound segment.

[0032] The target element corresponding to the target sliding window is in the first heart sound segment or the second heart sound segment of the sample heart sound signal. Under normal physiological conditions, the signal amplitude in the first heart sound segment is relatively higher than that in the second heart sound segment. Based on the neighborhood amplitude difference of adjacent target elements, it is determined whether the target element is in the first heart sound segment or the second heart sound segment. The specific method is as follows: Arrange the target elements in order in the amplitude sequence of each sample heart sound signal to obtain a target sequence; calculate the amplitude mean of all data points within the preset neighborhood range of the data points corresponding to each element in the target sequence in each sample heart sound signal as the neighborhood amplitude of the corresponding element; divide the target sequence into subsequences of length 2. If the neighborhood amplitude of the first element in the subsequence is greater than that of the second element, then mark the subsequence as a target subsequence; determine whether the number of target subsequences in each sample heart sound signal is greater than the number of non-target subsequences. If so, mark the time corresponding to the first element in the subsequences of the target sequence as the first heart sound time, and if not, mark the time corresponding to the second element in the subsequences of the target sequence as the second heart sound time.

[0033] The amplitude of the first heart sound segment is significantly positively correlated with myocardial contractility. In the case of decreased myocardial contractility in heart failure patients, the amplitude of the first heart sound segment may be lower than that of the second heart sound segment. It is impossible to accurately judge the first heart sound time and the second heart sound time only by the amplitude magnitudes of the first heart sound segment and the second heart sound segment. Although the signal peak within the second heart sound segment may be greater than the signal peak within the first signal segment, overall, due to the strong blood movement during the systolic phase of the heart and the relatively stable blood during the diastolic phase, the amplitude around the first heart sound time is greater than the amplitude around the second heart sound time.

[0034] In a specific implementation manner of the embodiment of the present invention, the result of rounding down the ratio obtained by taking the minimum value of the durations of the third heart sound segment and the fourth heart sound segment of all sample heart sound signals as the numerator and the length of the target sliding window as the denominator is used as the preset number.

[0035] In a specific implementation manner of the embodiment of the present invention, the data point is located at the middle position within its preset neighborhood range, and 11 data points are included within the preset neighborhood range of the data point.

[0036] Step S3: Denote the time period between two adjacent first heart sound moments of each sample heart sound signal as the cardiac cycle; determine the window combinations based on the position distributions of two types of sliding windows on each sample heart sound signal; fuse the feature vectors of the two sliding windows in each window combination according to the amplitude similarity between the short sliding window in each window combination of each sample heart sound signal and its neighboring cardiac cycles, and the distance between the short sliding window and its cardiac cycle to obtain the fusion vector of each window combination of each sample heart sound signal.

[0037] In order to perform multi-scale feature analysis on the sample heart sound signals, window combinations of different scales are constructed from the short sliding windows and long sliding windows of the sample heart sound signals. The specific method includes: for each sample heart sound signal, calculate the time intervals between the middle moments of each short sliding window on the sample heart sound signal and all long sliding windows respectively, and form a window combination by the long sliding window corresponding to the minimum time interval and each short sliding window.

[0038] The amplitude similarity between the cardiac cycle and its neighboring cardiac cycles presents the sensitivity degree of the distance of the sliding window pair within the cardiac cycle in the window combination, and in combination with the distance between the short sliding window in the window combination and its cardiac cycle, the feature vectors of the sliding windows of the window combination are adaptively fused, so that the fusion vector can better reflect the dynamic change process of the heart sound signal, which helps to more accurately identify the positions of various heart sounds in the cardiac cycle in the subsequent heart sound signal segmentation process.

[0039] Please refer to Figure 2 , which shows the flowchart of the steps of a method for obtaining a fusion vector provided by an embodiment of the present invention. The method includes: Step S310: Obtain the distance sensitivity of the sliding window within the corresponding cardiac cycle according to the amplitude similarity between each cardiac cycle of each sample heart sound signal and its neighboring cardiac cycles.

[0040] Preferably, in some possible implementation manners of the embodiments of the present invention, the method for obtaining the distance sensitivity includes: taking the sum of the absolute values of the amplitude differences between each sample heart sound signal and the start time of its neighboring cardiac cycles and the absolute values of the amplitude differences at the second heart sound time in each cardiac cycle as the overall neighborhood difference degree of each sample heart sound signal in each cardiac cycle; taking the ratio of the neighborhood amplitude of the amplitude at the start time of the cardiac cycle to the amplitude at the second heart sound time as the neighborhood ratio; taking the variance of the difference between each cardiac cycle of each sample heart sound signal and the neighborhood ratio of its neighboring cardiac cycles as the neighborhood discrete value of each cardiac cycle; and obtaining the distance sensitivity of the sliding window of each sample heart sound signal in each cardiac cycle according to the overall neighborhood difference degree and the neighborhood discrete value.

[0041] In patients with heart failure, the heart function is impaired, and there may be situations such as cardiac cycle disorders and changes in heart sound signal characteristics, resulting in different sensitivities of the sliding window to the distance of the first heart sound period in different cardiac cycles. If the amplitudes of each cardiac cycle of the sample heart sound signal are more similar to those of its neighboring cardiac cycles, it indicates that the characteristics of the sample heart sound samples are more stable in different cardiac cycles, and there is no need to capture the subtle differences caused by distance changes too acutely. Then, the sensitivity of the sliding window to the amplitude change at the first heart sound time in each cardiac cycle is smaller. Since the first heart sound time and the second heart sound time have unique and relatively stable physiological characteristics, the amplitude similarity between each sample heart sound signal and its neighboring cardiac cycles at the first cardiac moment and the second cardiac moment can be used to measure the amplitude similarity between each cardiac cycle and its neighboring cardiac cycles. That is, if the overall neighborhood difference degree and the neighborhood discrete value are smaller, the amplitude similarity between the cardiac cycle and its neighboring cardiac cycles is greater, and the distance sensitivity of the sliding window within the cardiac cycle is smaller.

[0042] In a specific implementation manner of the embodiments of the present invention, the distance sensitivity is expressed by the formula: ; In the formula, is the distance sensitivity of the sliding window in each cardiac cycle of each sample heart sound signal; is the neighborhood discrete value of each cardiac cycle of each sample heart sound signal; N is the total number of neighboring cardiac cycles of each cardiac cycle of each sample heart sound signal; F1 is the amplitude at the start time of each cardiac cycle of each sample heart sound signal: is the amplitude at the start time of the nth neighboring cardiac cycle of each cardiac cycle of each sample heart sound signal; is the amplitude at the second heart sound time of each cardiac cycle of each sample heart sound signal; is the amplitude at the second heart sound time of the nth neighboring cardiac cycle of each cardiac cycle of each sample heart sound signal; is the overall neighborhood difference degree of each cardiac cycle of each sample heart sound signal; is the absolute value function; Norm is the normalization function. It should be noted that there is only one second heart sound moment between each cardiac cycle.

[0043] In a specific implementation manner of the embodiment of the present invention, the adjacent 5 previous cardiac cycles and the adjacent 5 subsequent cardiac cycles of each cardiac cycle of the sample heart sound signal are used as the neighborhood cardiac cycles of each cardiac cycle.

[0044] Step S320: Denote the time interval between the middle moment in the short sliding window and the start moment in the cardiac cycle where the short sliding window is located in each window combination of each sample heart sound signal as the judgment distance of the corresponding window combination; use the distance sensitivity of the cardiac cycle where the short sliding window is located in each window combination of each sample heart sound signal to adjust the judgment distance of each window combination, and obtain the fusion coefficient of the short sliding window in the corresponding window combination.

[0045] Preferably, in some possible implementation manners of the embodiment of the present invention, the method for obtaining the fusion coefficient includes: using the distance sensitivity of the cardiac cycle where the short sliding window is located in each window combination of each sample heart sound signal to perform weighted processing on the judgment distance, and performing negative correlation and normalization processing on the weighted result to obtain the fusion coefficient of the short sliding window in the corresponding window combination.

[0046] The first heart sound moment is generated due to the sudden closure of the valve. The signal changes rapidly within the first heart sound segment, and the short sliding window has a high time resolution and can keenly capture the rapidly changing features in the heart sound signal. Then, within the cardiac cycle, the short sliding window closer to the first heart sound moment can better capture the key time point in the first heart sound segment, that is, the detailed information of the first heart sound moment, and a greater weight needs to be given to the short sliding window in the window combination. Although the long sliding window has a low time resolution, it can cover a longer time period of the heart sound signal. In the diastolic phase far from the first heart sound segment, the heart sound signal is relatively stable, and the long sliding window more shows the overall rhythm and low-frequency changes. Therefore, when the middle moment in the short sliding window in the window combination is closer to the start moment in its corresponding cardiac cycle, during the feature fusion process of the sliding window in the window combination, the weight of the short sliding window is greater, that is, the fusion coefficient is greater, and the weight of the long sliding window is smaller.

[0047] Use the sensitivity of the short sliding period in the window combination to the distance to adjust the judgment distance of the window combination, so that the adjusted distance can more accurately measure the detailed information of the cardiac cycle. In a specific implementation manner of the embodiment of the present invention, the fusion coefficient is expressed by the formula: ; In the formula, is the fusion coefficient of the a-th combined window of each sample heart sound signal; is the distance sensitivity of the sliding window within the cardiac cycle of the short sliding window in the a-th combined window of each sample heart sound signal; is the middle moment of the short sliding window in the a-th combined window of each sample heart sound signal; is the start moment of the cardiac cycle of the short sliding window in the a-th combined window of each sample heart sound signal; is the judgment distance of the a-th combined window of each sample heart sound signal; is the absolute value function; exp is the exponential function with the natural constant as the base.

[0048] Step S330: Use the fusion coefficient to fuse the feature vectors of the two sliding windows in each window combination of each sample heart sound signal to obtain the fusion vector of each window combination of each sample heart sound signal.

[0049] Preferably, the method for obtaining the fusion vector is: for each window combination of each sample heart sound signal, take the fusion coefficient as the weight of the feature vector of the short sliding window in the window combination, and the difference between the constant 1 and the fusion coefficient as the weight of the feature vector of the long sliding window in the window combination. Half of the result of the weighted sum of the feature vectors of the short sliding window and the long sliding window in the window combination is used as the fusion vector of the window combination.

[0050] ; In the formula, is the fusion vector of the a-th combined window of each sample heart sound signal; is the fusion coefficient of the a-th combined window of each sample heart sound signal; is the feature vector of the short sliding window in the a-th combined window of each sample heart sound signal; is the feature vector of the long sliding window in the a-th combined window of each sample heart sound signal. It should be noted that adding two vectors is equivalent to adding the elements in the same dimension of the two vectors respectively.

[0051] Step S4: Construct a hidden Markov model based on the fusion vectors of the window combinations of the sample heart sound signals, and use the hidden Markov model to divide the heart sound signal of the patient to be tested into different heart sound segments; perform heart sound abnormality detection on the patient to be tested based on the heart sound segments of the heart sound signal.

[0052] Arrange the fusion vectors of all windows of each sample heart sound signal in sequence to obtain the feature sequence of each sample heart sound signal; input the feature sequences of all sample heart sound signals into a hidden Markov model to construct the hidden Markov model, and the hidden Markov model divides the sample heart sound signals into four heart sound segments. Among them, the parameters in the model are estimated by the expectation maximization algorithm such as the forward-backward algorithm. The construction and training methods of the hidden Markov model are well-known techniques to those skilled in the art and will not be elaborated here.

[0053] In the present invention, the disease types of heart failure patients are labeled through a long short-term memory network. The input of the neural network is the sample heart sound signal with well-divided heart sound segments, and the output is the digital label of the disease type of the patient corresponding to the sample heart sound signal. The disease types in the embodiments of the present invention include acute heart failure and chronic heart failure, and the implementer can set them according to the actual situation. Among them, the relevant content of the long short-term memory network includes: the data set of the neural network is divided into a training set and a validation set; the training process of the neural network is the labeling process of the disease type. The specific labeling process is: label the sample heart sound signals of patients with acute heart failure as 0, and label the sample heart sound signals of patients with chronic heart failure as 1; the loss function of the neural network is the cross-entropy function. Among them, the long short-term memory network is a well-known technique to those skilled in the art and will not be elaborated here.

[0054] First, input the heart sound signal of the patient to be tested into the hidden Markov model, and output the heart sound signal with well-divided heart sound segments; then, input the heart sound signal of the patient to be tested into the trained neural network, and output the disease type of the patient to be tested; finally, initiate a targeted treatment plan for patients with acute heart failure.

[0055] It should be noted that the method for obtaining the feature sequence of the sample heart sound signal is the same as that of the heart sound signal of the patient to be tested.

[0056] So far, the present invention is completed.

[0057] Embodiment 2: The present invention also proposes a schematic diagram of a computer device for identifying abnormal heart sounds of patients with early acute heart failure. Please refer to Figure 3 Figure, the computer device includes a memory 501, a processor 502, and a computer program 503 stored in the memory 501 and running on the processor 502. Among them, when the processor 502 executes the computer program 503, the computer device can execute any one of the above-introduced methods for identifying abnormal heart sounds of patients with early acute heart failure.

[0058] In addition, an embodiment of the present application also protects a device, which may include a memory and a processor. Among them, executable program code is stored in the memory, and the processor is used to call and execute the executable program code to execute a method for identifying abnormal heart sounds of patients with early acute heart failure provided by an embodiment of the present application.

[0059] This embodiment can divide the functions of the device according to the above method examples. For example, it can correspond to each function module, or integrate two or more functions into one processing module. The above integrated modules can be implemented in the form of hardware. It should be noted that the division of modules in this embodiment is illustrative, only a logical function division, and there may be other division methods in actual implementation.

[0060] In the case of dividing each module according to each function, the device may further include a communication module, a signal analysis module, a complexity analysis module, a positioning module, etc. It should be noted that all relevant contents of each step involved in the above method embodiment can be cited in the function description of the corresponding function module, and will not be repeated here.

[0061] It should be understood that the device provided in this embodiment is used to execute the above method for identifying abnormal heart sounds of patients with early acute heart failure, so it can achieve the same effect as the above implementation method.

[0062] In the case of adopting an integrated unit, the device may include a processing module and a storage module. Among them, when the device is applied to a device, the processing module can be used to control and manage the actions of the device. The storage module can be used to support the device to execute mutual program code, etc.

[0063] Among them, the processing module can be a processor or a controller, which can implement or execute various exemplary logical blocks, modules and circuits included in the disclosure of the present application. The processor can also be a combination of computing functions, such as including a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc. The storage module can be a memory.

[0064] Embodiment 3: This embodiment also provides a computer-readable storage medium, in which computer program code is stored. When the computer program code runs on a computer, it causes the computer to execute the above related method steps to implement a method for identifying abnormal heart sounds of patients with early acute heart failure provided by the above embodiment.

[0065] Embodiment 4: This embodiment also provides a computer program product. When the computer program product runs on a computer, it causes the computer to execute the above-related steps to implement a method for identifying abnormal heart sounds in patients with early acute heart failure provided by the above embodiment.

[0066] Among them, the device, computer-readable storage medium, computer program product or chip provided in this embodiment are all used to execute the corresponding methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding methods provided above, and will not be elaborated here.

[0067] In the embodiments provided in this application, it should be understood that the disclosed device and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0068] It should be noted that the above sequence 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 drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0069] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments.

Claims

1. A method for identifying abnormal heart sounds in patients with early acute heart failure, characterized in that: The method includes: Acquire a heart sound signal of a patient to be tested and a sample heart sound signal, wherein the sample heart sound signal is divided into four heart sound segments; Use different types of sliding windows to traverse each sample heart sound signal, and extract the feature vector of each sliding window on the sample heart sound signal; according to the amplitude distribution and duration of different types of heart sound segments of each sample heart sound signal, obtain the first heart sound moment and the second heart sound moment of each sample heart sound signal; The time period between two adjacent first heart sound moments of each sample heart sound signal is recorded as a cardiac cycle; the window combination is determined based on the position distribution of the two sliding windows on each sample heart sound signal; according to the amplitude similarity between the cardiac cycle of the short sliding window and its neighboring cardiac cycle in each window combination of each sample heart sound signal, and the distance between the short sliding window and its cardiac cycle, the feature vectors of the two sliding windows in each window combination are fused to obtain a fusion vector of each window combination of each sample heart sound signal; A hidden Markov model is constructed based on the fusion vector of the window combination of the sample heart sound signal, and the heart sound signal of the patient to be tested is divided into different heart sound segments using the hidden Markov model; and heart sound abnormality detection is performed on the patient to be tested based on the heart sound segments of the heart sound signal.

2. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 1, characterized in that: The step of obtaining the first heart sound moment and the second heart sound moment of each sample heart sound signal includes: The types of heart sound segments include the first heart sound segment, the second heart sound segment, the third heart sound segment and the fourth heart sound segment; A target sliding window is set, wherein the length of the target sliding window is the minimum value of the duration of the first heart sound segment and the second heart sound segment of all sample heart sound signals; each sample heart sound signal is traversed using the target sliding window, and the peak values ​​of the signals in all target sliding windows on each sample heart sound signal are arranged in time sequence to obtain an amplitude sequence; For the remaining elements in the amplitude sequence except for a number of consecutive elements that are less than the judgment threshold, if each element is greater than its adjacent previous element, each element is recorded as a target element; the judgment threshold is equal to the amplitude mean of all sample points of the third heart sound segment and the fourth heart sound segment of all sample heart sound signals; According to the amplitude difference of data points within a preset neighborhood range of adjacent target element corresponding data points in the amplitude sequence of each sample heart sound signal, the moment corresponding to the target element in the amplitude sequence is divided into a first heart sound period and a second heart sound moment.

3. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 2, characterized in that: The step of dividing the time corresponding to the target element in the amplitude sequence into a first heart sound period and a second heart sound moment includes: Arranging the target elements in the amplitude sequence of each sample heart sound signal in order to obtain a target sequence; Calculate the amplitude mean of all data points in a preset neighborhood range of the corresponding data point in each sample heart sound signal for each element in the target sequence as the neighborhood amplitude of the corresponding element; The target sequence is divided into subsequences of length 2. If the neighborhood amplitude of the first element in the subsequence is greater than the neighborhood amplitude of the second element, the subsequence is recorded as a target subsequence. It is determined whether the number of target subsequences of each sample heart sound signal is greater than the number of non-target subsequences. If so, the time corresponding to the first element in the subsequence of the target sequence is recorded as the first heart sound time. Otherwise, the time corresponding to the second element in the subsequence of the target sequence is recorded as the second heart sound time.

4. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 3, characterized in that: The method for obtaining the fusion vector of each window combination of each sample heart sound signal includes: According to the amplitude similarity between each cardiac cycle of each sample heart sound signal and its neighboring cardiac cycle, the distance sensitivity of the sliding window in the corresponding cardiac cycle is obtained; The time interval between the middle moment in the short sliding window and the start moment in the cardiac cycle in which the short sliding window is located in each window combination of each sample heart sound signal is recorded as the judgment distance of the corresponding window combination; the judgment distance of each window combination is adjusted by using the distance sensitivity of the cardiac cycle in which the short sliding window is located in each window combination of each sample heart sound signal, and the fusion coefficient of the short sliding window in the corresponding window combination is obtained; The fusion coefficient is used to fuse the feature vectors of the two sliding windows in each window combination of each sample heart sound signal to obtain a fusion vector of each window combination of each sample heart sound signal.

5. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 4, characterized in that: The method for obtaining the distance sensitivity includes: The sum of the absolute value of the amplitude difference between each sample heart sound signal at the beginning of each cardiac cycle and the absolute value of the amplitude difference at the second heart sound moment of its neighboring cardiac cycle is taken as the overall neighborhood difference degree of each sample heart sound signal in each cardiac cycle; The ratio of the neighborhood amplitude between the start time of the cardiac cycle and the amplitude of the second heart sound is taken as the neighborhood ratio; the variance of the difference between each cardiac cycle of each sample heart sound signal and the neighborhood ratio of its neighboring cardiac cycle is recorded as the neighborhood discrete value of each cardiac cycle; According to the overall neighborhood difference and the neighborhood discrete value, the distance sensitivity of the sliding window of each sample heart sound signal in each cardiac cycle is obtained.

6. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 4, characterized in that: The method for obtaining the fusion coefficient includes: The distance sensitivity of the cardiac cycle in which the short sliding window in each window combination of each sample heart sound signal is located is used to perform weighted processing on the judgment distance, and the weighted result is negatively correlated and normalized to obtain the fusion coefficient of the short sliding window in the corresponding window combination.

7. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 4, characterized in that: The fusion vector of each window combination of each sample heart sound signal includes: For each window combination of each sample heart sound signal, the fusion coefficient is used as the weight of the feature vector of the short sliding window in the window combination, the difference between the constant 1 and the fusion coefficient is used as the weight of the feature vector of the long sliding window in the window combination, and half of the result of the weighted sum of the feature vectors of the short sliding window and the long sliding window in the window combination is used as the fusion vector of the window combination.

8. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 1, characterized in that: The method of constructing a hidden Markov model based on the fusion vector of the window combination of the sample heart sound signal includes: The fusion vectors of all window combinations of each sample heart sound signal are arranged in sequence to obtain a feature sequence of each sample heart sound signal; the feature sequences of all sample heart sound signals are input into a hidden Markov model to construct a hidden Markov model.

9. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 1, characterized in that: The determining of the window combination based on the position distribution of the two sliding windows on each sample heart sound signal includes: For each sample heart sound signal, the time interval between each short sliding window and the middle moment of all long sliding windows on the sample heart sound signal is calculated, and a window combination is formed by the long sliding window corresponding to the smallest time interval and each short sliding window.

10. The method for identifying abnormal heart sounds in patients with early-stage acute heart failure according to claim 1, characterized in that: The types of sliding windows include: long sliding windows and short sliding windows.

Images