CN111370120B - Heart diastole dysfunction detection method based on heart sound signals - Google Patents
Heart diastole dysfunction detection method based on heart sound signals Download PDFInfo
- Publication number
- CN111370120B CN111370120B CN202010097640.4A CN202010097640A CN111370120B CN 111370120 B CN111370120 B CN 111370120B CN 202010097640 A CN202010097640 A CN 202010097640A CN 111370120 B CN111370120 B CN 111370120B
- Authority
- CN
- China
- Prior art keywords
- heart sound
- detected
- sound signal
- diastolic dysfunction
- mel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B7/00—Instruments for auscultation
- A61B7/02—Stethoscopes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/18—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being spectral information of each sub-band
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/24—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being the cepstrum
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
- G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/51—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination
- G10L25/66—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for comparison or discrimination for extracting parameters related to health condition
Abstract
The invention discloses a heart sound signal-based detection method for diastolic dysfunction, which is characterized in that the heart sound signal-based detection method is characterized in that the Mel frequency cepstrum coefficient of a heart sound signal to be detected is extracted; and inputting the Mel frequency cepstrum coefficient into a trained detection network model, and outputting a diastolic dysfunction category corresponding to the heart sound signal to be detected through the detection network model. The invention converts the heart sound signals into two-dimensional time-frequency characteristics by utilizing the neural network, and determines the category of diastolic dysfunction corresponding to the heart sound signals according to the two-dimensional time-frequency characteristics, thereby rapidly and conveniently completing heart sound signal detection. Meanwhile, the invention can be integrated into terminal equipment, and provides a certain clinical reference for doctors in places such as community hospitals, village and town hospitals and the like which lack professional doctor resources.
Description
Technical Field
The invention relates to the technical field of biological signal processing, in particular to a heart diastole dysfunction detection method based on heart sound signals.
Background
Diastolic dysfunction often results from impaired filling of the ventricles, resulting in reduced ventricular end-diastole volume, elevated end-diastole pressure, or both. Diastolic dysfunction can be seen in any age group of people, even young children, especially children with heart defects at birth, which is more common in the elderly. About 2300 million people worldwide suffer from this disease. Diastolic dysfunction is also a cause of heart failure, and it is incompletely estimated that about 50% of heart failure patients are caused by diastolic dysfunction. At present, echocardiography examination is an important examination method for assessing diastolic dysfunction.
Heart sound is one of important physiological signals of a human body, and is a composite sound generated by opening and closing of heart valves, diastole and contraction of tendons and muscles, impact of blood flow, and vibration of heart vessel walls. Clinically, auscultation of heart sounds is a basic method of evaluating the heart, and can be performed by means of murmurs and distortions occurring in heart sounds as important diagnostic information. The diagnosis and analysis of heart sounds are noninvasive and convenient methods for knowing the functional states of heart and blood vessels, the auscultation effect of the traditional doctor mainly depends on the clinical experience and subjective judgment of the doctor, and the pathological information of heart sounds cannot be recorded for reference for the subsequent diagnosis and analysis of heart sounds, so that objectivity and subjectivity are lacked. For modern information processing technology, the extracted heart sounds can be analyzed and diagnosed to obtain the health condition of the heart through simple heart sound acquisition, but due to the fact that noise exists in the acquisition process, the heart sound signals are generally obvious in commonality, the identification degree is not high, and the like, the heart sound analysis has certain difficulty in practical application.
Disclosure of Invention
The invention aims to solve the technical problem of providing a heart diastole dysfunction detection method based on heart sound signals aiming at the defects of the prior art.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows:
a method of detecting diastolic dysfunction based on heart sound signals, the method comprising:
extracting a mel frequency cepstrum coefficient of a heart sound signal to be detected, wherein the heart sound signal comprises a plurality of cardiac cycles;
and inputting the Mel frequency cepstrum coefficient into a trained detection network model, and outputting a diastolic dysfunction category corresponding to the heart sound signal to be detected through the detection network model.
The method for detecting diastolic dysfunction based on the heart sound signal, wherein before the extracting of the mel frequency cepstrum coefficient of the heart sound signal to be detected, the method further comprises:
acquiring a heart sound signal, and randomly selecting a preset number of heart sound fragments with equal duration from the heart sound signal, wherein each heart sound fragment comprises at least one cardiac cycle;
each heart sound segment is used as a heart sound signal to be detected, and the mel frequency cepstrum coefficient operation for extracting the heart sound signal to be detected is performed for each heart sound segment.
The method for detecting the diastolic dysfunction based on the heart sound signals, wherein after outputting the diastolic dysfunction category corresponding to the heart sound signals to be detected through the detection network model, further comprises:
and acquiring the diastolic dysfunction category corresponding to each heart sound fragment, and determining the diastolic dysfunction category corresponding to the heart sound signal based on all the acquired diastolic dysfunction categories.
The heart sound signal-based detection method for diastolic dysfunction, wherein the mel-frequency cepstrum coefficient comprises: static mel frequency cepstral coefficients and first order dynamic differential parameters of the static mel frequency cepstral coefficients.
The heart sound signal-based detection method for diastolic dysfunction, wherein the extracting of mel frequency cepstrum coefficients of the heart sound signal to be detected specifically comprises:
performing fast Fourier transform on the heart sound signal to be detected to obtain a frequency spectrum sequence corresponding to the heart sound signal to be detected;
performing Mel scale filtering on the frequency spectrum sequence, and calculating the logarithmic energy of the heart sound signal to be detected according to the frequency spectrum sequence after Mel scale filtering;
determining a static Mel frequency cepstrum coefficient corresponding to the heart sound signal to be detected according to the logarithmic energy;
performing first-order difference on the static Mel frequency cepstrum coefficient to obtain a first-order dynamic difference parameter corresponding to the heart sound signal to be detected;
and obtaining the Mel frequency cepstrum coefficient of the heart sound signal to be detected according to the static Mel frequency cepstrum coefficient and the first-order dynamic difference parameter.
The method for detecting diastolic dysfunction based on heart sound signals, wherein before performing fast fourier transform on the heart sound signals to be detected to obtain a spectrum sequence corresponding to the heart sound signals to be detected, the method further comprises:
filtering the heart sound signal to be detected, and framing the filtered information of the signal to be detected to obtain a heart sound frame sequence corresponding to the heart sound signal to be detected;
and carrying out windowing on each heart sound frame in the frame sequence, and taking the heart sound frame sequence subjected to windowing as a heart sound signal to be detected.
The heart sound signal-based heart diastolic dysfunction detection method is characterized in that the number of corresponding heart sound sampling points of any two heart sound frames in the heart sound frame sequence is the same, overlapping exists between any two adjacent heart sound frames, and the number of the overlapped heart sound sampling points is the same.
The heart sound signal-based detection method for diastolic dysfunction, wherein the detection network model comprises a residual error module, a cyclic convolution module and a full connection layer; inputting the mel-frequency cepstrum coefficient into a trained detection network model, and outputting a diastolic dysfunction category corresponding to a heart sound signal to be detected through the detection network model specifically comprises:
inputting the mel frequency cepstrum coefficient into the residual module, and outputting a specific spectrum characteristic through the residual module;
inputting the specific spectrum characteristics into the cyclic convolution module, and outputting time domain strengthening characteristics of the specific spectrum characteristics through the cyclic convolution module;
outputting the time domain strengthening characteristic of the specific frequency spectrum characteristic to the full-connection layer, and outputting the diastolic dysfunction category corresponding to the heart sound signal to be detected through the full-connection layer.
A computer readable storage medium storing one or more programs executable by one or more processors to implement steps in a method of detecting diastolic dysfunction based on heart sound signals as described in any of the above.
A terminal device, comprising: a processor, a memory, and a communication bus; the memory has stored thereon a computer readable program executable by the processor;
the communication bus realizes connection communication between the processor and the memory;
the processor, when executing the computer readable program, implements the steps in a method for detecting diastolic dysfunction based on heart sound signals as described in any of the above.
The beneficial effects are that: compared with the prior art, the invention provides a heart sound signal-based detection method for diastolic dysfunction, which is characterized in that the heart sound signal-based detection method is implemented by extracting the mel frequency cepstrum coefficient of the heart sound signal to be detected; and inputting the Mel frequency cepstrum coefficient into a trained detection network model, and outputting a diastolic dysfunction category corresponding to the heart sound signal to be detected through the detection network model. The invention converts the heart sound signals into two-dimensional time-frequency characteristics by utilizing the neural network, and determines the category of diastolic dysfunction corresponding to the heart sound signals according to the two-dimensional time-frequency characteristics, thereby rapidly and conveniently completing heart sound signal detection. Meanwhile, the invention can be integrated into terminal equipment, and provides a certain clinical reference for doctors in places such as community hospitals, village and town hospitals and the like which lack professional doctor resources.
Drawings
Fig. 1 is a flowchart of a method for detecting diastolic dysfunction based on heart sound signals provided by the invention.
Fig. 2 is a schematic flow chart of a procedure for extracting mel frequency cepstrum coefficient according to the method for detecting diastolic dysfunction based on heart sound signals.
Fig. 3 is a schematic flow chart of a method for detecting diastolic dysfunction based on heart sound signals according to the present invention.
Fig. 4 is a schematic structural diagram of a residual module in the method for detecting diastolic dysfunction based on heart sound signals.
Fig. 5 is a schematic structural diagram of a convolutional network module in the method for detecting diastolic dysfunction based on heart sound signals provided by the invention.
Fig. 6 is a schematic structural diagram of a terminal device provided by the present invention.
Detailed Description
The invention provides a heart sound signal-based detection method for diastolic dysfunction, which aims to make the purposes, technical schemes and effects of the invention clearer and more definite, and further details the invention by referring to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. The term "and/or" as used herein includes all or any element and all combination of one or more of the associated listed items.
It will be understood by those skilled in the art that all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs unless defined otherwise. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The present embodiment provides a method for detecting diastolic dysfunction based on a heart sound signal, which can be applied to an electronic device that can be implemented in various forms. Such as a cell phone, tablet, palm top, personal digital assistant (Personal Digital Assistant, PDA), etc. In addition, the functions performed by the method may be performed by a processor in an electronic device that includes at least a processor and a storage medium, although the program code may be stored in a computer storage medium.
As shown in fig. 1, the present embodiment provides a method for detecting diastolic dysfunction based on heart sound signals, which may include the steps of:
s10, extracting a Mel frequency cepstrum coefficient of a heart sound signal to be detected, wherein the heart sound signal comprises a plurality of cardiac cycles.
Specifically, the heart sound is one of important physiological signals of a human body, and is a composite sound generated by opening and closing of a heart valve, diastole and contraction of tendons and muscles, impact of blood flow and vibration of a heart vessel wall. The heart sound signal to be detected may be a heart sound signal collected by a heart sound collection device, may be a heart sound signal sent by an external device, may also be a heart sound signal obtained through the internet (for example, hundred degrees, etc.), and the like. For example, the heart sound signal is acquired by a heart sound acquisition device, the sampling frequency of the heart sound signal is 4000Hz, and the heart sound signal comprises several cardiac cycles. The Mel-frequency cepstrum coefficient is a coefficient constituting a Mel-frequency cepstrum, wherein the Mel-frequency cepstrum (Mel-Frequency Cepstrum) is a linear transformation of a logarithmic energy spectrum based on a nonlinear Mel scale (Mel scale) of sound frequencies. In addition, in the case of the optical fiber,
further, in order to avoid the chance of the heart sound signal, in an implementation manner of the present embodiment, before the extracting the mel-frequency cepstrum coefficient of the heart sound signal to be detected, the method further includes:
s01, acquiring heart sound signals, and randomly selecting a preset number of heart sound fragments with equal duration from the heart sound signals, wherein each heart sound fragment comprises at least one cardiac cycle;
and S02, taking each heart sound segment as a heart sound signal to be detected, and executing the operation of extracting the mel frequency cepstrum coefficient of the heart sound signal to be detected for each heart sound segment.
Specifically, the preset number of heart sound segments are all partial heart sound signals extracted from the heart sound signals, and each heart sound segment comprises at least one complete cardiac cycle, wherein the heart sound duration corresponding to any two heart sound segments in the preset number of heart sound segments is equal, and the heart sound duration corresponding to the heart sound segments is longer than the cycle duration of the cardiac cycle. In addition, the heart sound signals corresponding to any two heart sound segments are not identical, and it can be understood that, for any heart sound segment in the preset number of heart sound segments, the collection time of the heart sound signal corresponding to the heart sound segment may be partially overlapped with or may not be overlapped with the collection time corresponding to any heart sound segment except for the heart sound segment in the preset number of heart sound segments, but the overlapping cannot be completed. In addition, the preset number of heart sound segments can be obtained by randomly selecting and dividing the heart sound signal for a preset number of times, and for each random selection, when the heart sound segments are selected, whether the selected heart sound segments are overlapped with the heart sound segments already selected or not can be judged, if not, the heart sound segments are saved, and if so, the heart sound segments are discarded.
Illustrating: assuming that the heart sound signal is a heart sound signal with a length of 30 seconds and the preset number is 10, 10 random selection divisions with a duration of 3 seconds are performed on the heart sound signal with a length of 30 seconds, so as to select 10 heart sound fragments from the information signal with a length of 30 seconds, so that each heart sound fragment with a duration of 3 seconds can contain a complete cardiac cycle. This is because the normal range of the cardiac cycle is around 0.8 seconds, whereas the heart sound duration of the heart sound segment is 3 seconds, so that the heart sound segment may contain the complete cardiac cycle.
Further, in an embodiment of the present embodiment, the mel-frequency cepstrum coefficient includes: the static Mel frequency cepstrum coefficient and the first-order dynamic differential parameter of the static Mel frequency cepstrum coefficient can better embody the time domain continuity of the heart sound signal through the structure of the static Mel frequency cepstrum coefficient and the first-order dynamic differential parameter of the static Mel frequency cepstrum coefficient. Correspondingly, as shown in fig. 2, the extracting mel-frequency cepstrum coefficients of the heart sound signal to be detected specifically includes:
s21, performing fast Fourier transform on the heart sound signal to be detected to obtain a frequency spectrum sequence corresponding to the heart sound signal to be detected;
s22, carrying out Mel scale filtering on the spectrum sequence, and calculating the logarithmic energy of the heart sound signal to be detected according to the spectrum sequence after Mel scale filtering;
s23, determining a static Mel frequency cepstrum coefficient corresponding to the heart sound signal to be detected according to the logarithmic energy;
s24, carrying out first order difference on the static Mel frequency cepstrum coefficient to obtain a first order dynamic difference parameter corresponding to the heart sound signal to be detected;
s25, obtaining the Mel frequency cepstrum coefficient of the heart sound signal to be detected according to the static Mel frequency cepstrum coefficient and the first-order dynamic difference parameter.
Specifically, in the step S21, the fast fourier transform is performed on each of the heart sound frames included in the heart sound signal, so as to obtain a spectral distribution corresponding to each of the heart sound frames. In one implementation of this embodiment, the fast fourier transform may be expressed as:
where x (n) represents a heart sound frame, H represents the number of fourier transform points, k is the kth spectrum after fourier transform, and n represents the frame number of the heart sound frame.
Further, the heart sound frame is obtained by preprocessing a heart sound signal to be detected, as shown in fig. 2, the preprocessing process may include:
filtering the heart sound signal to be detected, and framing the filtered information of the signal to be detected to obtain a heart sound frame sequence corresponding to the heart sound signal to be detected;
and carrying out windowing on each heart sound frame in the frame sequence, and taking the heart sound frame sequence subjected to windowing as a heart sound signal to be detected.
Specifically, the filtering the heart sound signal to be detected is performed to perform high-pass filtering on the heart sound signal to be detected, it can be understood that, for the heart sound signal to be detected formed by each heart sound segment, the heart sound signal x (t) to be detected is passed through a high-pass filter to perform high-pass filtering on the heart sound signal to be detected by a high-pass filter degree, where the high-pass filter can be expressed as: h (z) =1- μz -1 And the mu is a filtering parameter and is used for determining the cut-off frequency and the bandwidth of the high-pass filter. Thus, when the heart sound signal to be detected is subjected to high-pass filtering, the heart sound signal to be detected and the high-pass filter are subjected to convolution operation to obtain a high-frequency enhanced heart sound signal x (t)', wherein,furthermore, in one implementation of this embodiment, the filtering parameter μmay be 0.97.
Further, the framing process divides the heart sound signal to be detected after the high frequency enhancement into a plurality of heart sound frames to obtain heart soundsA sequence of frames. The number of corresponding heart sound sampling points of any two heart sound frames in the heart sound frame sequence is the same, and overlapping exists between any two adjacent heart sound frames, and the number of the overlapped heart sound sampling points is the same. For example, 550 sampling points of the heart sound signal x (t) ' to be detected after high-frequency enhancement are integrated into one frame (the duration is about 25 milliseconds), 220 sampling points are overlapped between two adjacent frames (the duration is about 10 milliseconds), and each heart sound frame x (t) ' is obtained ' n Where n represents the frame number of the heart sound frame.
Further, the windowing process refers to multiplying each heart sound frame by a hamming window to obtain windowed heart sound frames, which can improve the continuity of the left and right ends of each heart sound frame in the heart sound frame sequence and can prevent spectrum leakage. For example, for the heart sound signal x (t) to be detected after high frequency enhancement' n Multiplying the high-frequency enhanced heart sound signal to be detected by a hamming window W (n) to obtain a windowed heart sound frame x (n), i.e., x (n) =x (t)' n X W (N), n=0, 1,..n-1, N is the number of frames of the heart sound frame sequence. Wherein, the hamming window may be expressed as:
where N represents the ordinal number of the frame, N is the number of frames of the heart sound frame sequence, α is the attenuation, and is used to determine the attenuation degree of the side lobe, for example, the value of α is 0.46.
Further, in the step S22, the Mel-scale filtering is to filter and scale-convert each frame spectrum after FFT by a Mel filter bank (noted Mel filter bank) to convert each frame spectrum after FFT by FFT to a Mel spectrum (noted Mel spectrum). In this embodiment, the Mel filter bank is a group of M-order triangular filters H (k), where the number of triangular filters is M, and the selection of M is determined according to the cut-off frequency or critical bandwidth of the heart sound signal. In the analysis of the heart sound signal, the effective frequency range of the heart sound signal is below 1000Hz, so that the number of triangular filters can be selected to be 12, i.e. m=12. In addition, the calculation formula corresponding to the logarithmic energy s (m) may be:
wherein M is the number of triangular filters, H m (k) Represents the mth triangular filter and X (k) represents the kth spectrum after the fast fourier transform.
Further, in the step S23, since two adjacent triangular filters have overlapping portions, there is a correlation between the energies output by the respective filter banks, so that in order to remove the correlation, discrete cosine transform may be performed on the log energy, and the correlation between the energies output by the respective filter banks may be removed by discrete cosine transform, so as to obtain a static mel frequency cepstrum coefficient. In addition, the high-frequency component of the signal can be shifted to low frequency through discrete cosine transform, so that the signal energy is concentrated at low frequency, and the energy concentration characteristic is improved. In this embodiment, the discrete cosine transform can be expressed as:
where C (n) is a parameter of the mel-frequency cepstrum coefficient, n=1, 2,..l, L is an order of the mel-frequency cepstrum coefficient, for example, l=40, m is the number of triangular filters, and s (m) is logarithmic energy.
Further, in the step S24, the static mel frequency cepstrum coefficient is subjected to first order difference to obtain a differential spectrum of dynamic characteristics, where the first order dynamic differential parameter calculation formula may be
Wherein d t Representing a first order difference of a t-th frame; c t+1 Representing the cepstral coefficient, c, of the t+1st frame t-1 And the cepstrum coefficient of the t-1 th frame is represented, and N is the number of frames of the heart sound frame sequence.
S20, inputting the Mel frequency cepstrum coefficient into a trained detection network model, and outputting a category of diastolic dysfunction corresponding to the heart sound signal to be detected through the detection network model.
Specifically, the detection network model is a deep neural network model, an input item of the detection network model is a mel frequency cepstrum coefficient corresponding to a heart sound signal, and an output item of the detection network model is a diastolic dysfunction category corresponding to the heart sound signal. It will be appreciated that, when the mel-frequency cepstrum coefficient is input into the detection network model, the detection network model may output the diastolic dysfunction category corresponding to the mel-frequency cepstrum coefficient, i.e. output the diastolic dysfunction category corresponding to the heart sound signal. In this embodiment, the detection network model includes a residual module, a cyclic convolution module, and a full connection layer; the residual error module is connected with the circular convolution module, the circular convolution module is connected with the full connection layer, wherein the residual error module is used for extracting specific frequency spectrum characteristics corresponding to the mel frequency cepstrum coefficient, the circular convolution module is used for extracting time domain strengthening characteristics, and the full connection layer is used for outputting the category of diastolic dysfunction.
Further, the detection network model is obtained through iterative learning training, wherein a residual error module is used for extracting specific spectrum features of the mel frequency cepstrum coefficient, reducing the resolution of a feature map of the mel frequency cepstrum coefficient, equally dividing the obtained feature map along a time dimension to obtain a submatrix of the feature matrix, sequentially inputting the submatrix into a circular convolution module for convolution, inputting a convolved result into a full-connection layer to obtain a specific feature vector, and carrying out final classification prediction; by comparing the heart sound label with the heart sound label, the classification error rate is reduced continuously and iteratively, a model with higher accuracy is finally learned, and the trained model is used for diagnosing heart sound of a user.
In one implementation manner of this embodiment, as shown in fig. 3, the inputting the mel-frequency cepstrum coefficient into a trained detection network model, and outputting, by the detection network model, a diastolic dysfunction category corresponding to a heart sound signal to be detected specifically includes:
s31, inputting the Mel frequency cepstrum coefficient into the residual error module, and outputting specific frequency spectrum characteristics through the residual error module;
s32, inputting the specific spectrum characteristics into the cyclic convolution module, and outputting time domain strengthening characteristics of the specific spectrum characteristics through the cyclic convolution module;
and S33, outputting the time domain strengthening characteristic of the specific frequency spectrum characteristic to the full-connection layer, and outputting the diastolic dysfunction category corresponding to the heart sound signal to be detected through the full-connection layer.
Specifically, in the step S31, the residual module may be a residual neural network, and after the mel frequency cepstrum coefficient M of the heart sound signal to be detected is input into the residual neural network, the residual neural network extracts the correlation of the time part and the correlation of the frequency domain part in the two-dimensional spectrum feature map through a multi-layer convolution operation, so as to obtain the specific spectrum feature f h,w In addition, the residual module can comprise 3 residual units, each residual unit comprises a convolution layer, a batch standardization layer and residual connection, and the residual convolution neural network extracts the correlation of the time part and the correlation of the frequency domain part in the two-dimensional spectrum characteristic diagram through multi-layer convolution operation. For example, in one implementation manner of the present embodiment, as shown in fig. 4, the residual neural network includes a first residual unit, a second residual unit, and a third residual unit that are sequentially connected, where the first residual unit is connected with the second residual unit in a residual manner, and the second residual unit is connected with the third residual unit in a residual manner; the first residual error unit comprises two serially connected convolution subunits, wherein each convolution subunit comprises a serially connected convolution layer, a nonlinear activation layer and a batch standardization layer; and the second residual error unit and the third residual error unit are structured, and the second residual error unit is formed by arranging a pooling convolution layer after the last layer of the first residual error unit. Furthermore, the neural network trains the specific spectral feature extraction network with a learning rate of 0.0001 using an Adam optimizer, resulting in a specific from the residual convolution neural networkSpectral features f h,w 。
Further, in the step S32, the cyclic convolution module may be a cyclic convolution neural network, and the information transfer is performed on the sequence feature picture through the cyclic convolution neural network, so as to obtain a time domain enhancement feature. Before inputting the specific spectrum characteristic into a circular convolution module, extracting the extracted specific spectrum characteristic f h,w Equally dividing into sequential features along the length of time And sequentially inputting the sequence features obtained by division into a cyclic convolution module. In one implementation manner of this embodiment, as shown in fig. 5, the cyclic convolution module feature segmentation unit, the cyclic convolution unit and the full connection layer, where the cyclic convolution unit includes three long short time memory networks LSTM, which are respectively recorded as a first long short time memory network, a second long short time memory network and a third long short time memory network, and the weights of the three cyclic convolution units are shared. In addition, the convolutional network module may be trained using Adam optimizers at a learning rate of 0.0001, and the convolutional network module and the residual module may be trained synchronously or separately.
Further, the cyclic convolution module transmits information to the sequence characteristic spectrum,output as H through the first long-short-time memory network 1 The hidden state of the first long-short-time memory network is transmitted as input to the hidden state of the second long-short-time memory network, and the output of the second long-short-time memory network is H 2 The method comprises the steps of carrying out a first treatment on the surface of the The hidden state of the second long-short-time memory network is transmitted as input to the hidden state of the third long-short-time memory network, and the output of the third long-short-time memory network is H 3 Thus, the time domain strengthening of the specific frequency spectrum characteristic is realized, and the obtained time domain strengthening result is H={H 1 ,H 2 ,H 3 }。
Further, in the step S23, a time domain enhancement feature h= { H corresponding to the specific spectrum feature 1 ，H 2 ,H 3 Input to the full connection layer to obtain a specificity characterization vector I= { a 1 ,a 2 }. The full-connection layer classifies the specific characterization vectors by using a Softmax function to obtain the category of the diastolic dysfunction corresponding to the heart sound signal to be detected. Wherein, the Softmax function may be expressed as:
wherein a is k Is the kth eigenvector value in the token vector.
Further, in one implementation manner of this embodiment, when the heart sound signal is acquired, a preset number of heart sound segments are selected from the heart sound signal, and each selected heart sound segment is used as the heart sound signal to be detected, so that a preset number of diastolic dysfunction categories can be acquired. Thus, when the diastolic dysfunction category corresponding to the heart sound signal is determined, the diastolic dysfunction category may be determined according to the acquired preset number. Correspondingly, the method further comprises the steps of: and acquiring the diastolic dysfunction category corresponding to each heart sound fragment, and determining the diastolic dysfunction category corresponding to the heart sound signal based on all the acquired diastolic dysfunction categories. Therefore, the characteristic information of a plurality of cardiac cycles of the patient is fully considered by the category of the diastolic dysfunction obtained through a plurality of heart sound fragments, the predicted contingency is reduced, and the predicted result is more accurate.
In order to further explain the method for detecting diastolic dysfunction based on the heart sound signal provided in this embodiment, a specific experimental example is given below. In this experimental example, the heart sound data used contains 38 heart sound data without diastolic dysfunction and 87 heart sound data with diastolic dysfunction, the time length of each heart sound data is 30 seconds, the heart sound records of 125 patients in 2 categories of whether the heart diastolic dysfunction exists are divided, an experiment is performed by adopting a 5-fold cross-validation method, the distribution of the categories in each fold is ensured to be the same as the distribution of the categories in the original data, 25 patients are used as a test set, 10 patients are divided as a verification set in 100 patients, and the rest are used as a training set. The segment-level diastolic dysfunction category is obtained through the convolutional neural network, 79.17% of accuracy, 82.35% of sensitivity and 71.43% of specificity can be obtained on the test set, and the method has important potential application in the aspect of diastolic dysfunction screening of the heart.
Based on the above-described method for detecting diastolic dysfunction based on heart sound signals, the present embodiment provides a computer-readable storage medium storing one or more programs executable by one or more processors to implement the steps in the method for detecting diastolic dysfunction based on heart sound signals as described in the above-described embodiments.
Based on the above-mentioned method for detecting diastolic dysfunction based on heart sound signals, the present invention also provides a terminal device, as shown in fig. 6, which includes at least one processor (processor) 20; a display screen 21; and a memory (memory) 22, which may also include a communication interface (Communications Interface) 23 and a bus 24. Wherein the processor 20, the display 21, the memory 22 and the communication interface 23 may communicate with each other via a bus 24. The display screen 21 is configured to display a user guidance interface preset in the initial setting mode. The communication interface 23 may transmit information. The processor 20 may invoke logic instructions in the memory 22 to perform the methods of the embodiments described above.
Further, the logic instructions in the memory 22 described above may be implemented in the form of software functional units and stored in a computer readable storage medium when sold or used as a stand alone product.
The memory 22, as a computer readable storage medium, may be configured to store a software program, a computer executable program, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 performs functional applications and data processing, i.e. implements the methods of the embodiments described above, by running software programs, instructions or modules stored in the memory 22.
The memory 22 may include a storage program area that may store an operating system, at least one application program required for functions, and a storage data area; the storage data area may store data created according to the use of the terminal device, etc. In addition, the memory 22 may include high-speed random access memory, and may also include nonvolatile memory. For example, a plurality of media capable of storing program codes such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or a transitory storage medium may be used.
In addition, the specific processes that the storage medium and the plurality of instruction processors in the terminal device load and execute are described in detail in the above method, and are not stated here.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (9)
1. A method for detecting diastolic dysfunction based on heart sound signals, the method comprising:
extracting a mel frequency cepstrum coefficient of a heart sound signal to be detected, wherein the heart sound signal comprises a plurality of cardiac cycles;
inputting the mel frequency cepstrum coefficient into a trained detection network model, and outputting a heart diastolic dysfunction category corresponding to a heart sound signal to be detected through the detection network model;
the detection network model comprises a residual error module, a cyclic convolution module and a full connection layer; inputting the mel-frequency cepstrum coefficient into a trained detection network model, and outputting a diastolic dysfunction category corresponding to a heart sound signal to be detected through the detection network model specifically comprises:
inputting the mel frequency cepstrum coefficient into the residual module, and outputting a specific spectrum characteristic through the residual module;
inputting the specific spectrum characteristics into the cyclic convolution module, and outputting time domain strengthening characteristics of the specific spectrum characteristics through the cyclic convolution module;
outputting the time domain strengthening characteristic of the specific frequency spectrum characteristic to the full-connection layer, and outputting a diastolic dysfunction category corresponding to the heart sound signal to be detected through the full-connection layer;
the step of inputting the mel-frequency cepstrum coefficient into the residual module, and the step of outputting the specific spectrum characteristic through the residual module specifically comprises:
the residual error module is a residual error neural network, the mel frequency cepstrum coefficient of the heart sound signal to be detected is input into the residual error neural network, and the residual error neural network extracts the correlation of time part and the correlation of frequency domain part in a two-dimensional spectrum characteristic diagram through multi-layer convolution operation to obtain the specific spectrum characteristic;
the step of inputting the specific spectrum characteristic into the cyclic convolution module, and the step of outputting the time domain strengthening characteristic of the specific spectrum characteristic through the cyclic convolution module specifically comprises the following steps:
the cyclic convolution module is a cyclic convolution neural network, the specific spectrum features are equally divided into three sequence features along the time length direction, the sequence features obtained by division are sequentially input into the cyclic convolution neural network, and the information transmission is carried out on the sequence features through the cyclic convolution neural network, so that time domain strengthening features are obtained;
outputting the time domain strengthening characteristic of the specific frequency spectrum characteristic to the full-connection layer, and outputting the diastolic dysfunction category corresponding to the heart sound signal to be detected through the full-connection layer specifically comprises:
and inputting the time domain strengthening characteristic into a full-connection layer to obtain a specificity characterization vector, classifying the specificity characterization vector by the full-connection layer, and outputting the type of the diastolic dysfunction corresponding to the heart sound signal to be detected.
2. The method for detecting diastolic dysfunction based on heart sound signals according to claim 1, wherein before extracting mel-frequency cepstrum coefficients of the heart sound signals to be detected, the method further comprises:
acquiring a heart sound signal, and randomly selecting a preset number of heart sound fragments with equal duration from the heart sound signal, wherein each heart sound fragment comprises at least one cardiac cycle;
each heart sound segment is used as a heart sound signal to be detected, and the mel frequency cepstrum coefficient operation for extracting the heart sound signal to be detected is performed for each heart sound segment.
3. The method for detecting diastolic dysfunction based on heart sound signals according to claim 2, wherein after outputting the class of diastolic dysfunction corresponding to the heart sound signals to be detected through the detection network model, further comprises:
and acquiring the diastolic dysfunction category corresponding to each heart sound fragment, and determining the diastolic dysfunction category corresponding to the heart sound signal based on all the acquired diastolic dysfunction categories.
4. The method for detecting diastolic dysfunction based on heart sound signals according to claim 1, wherein the mel-frequency cepstrum coefficient includes: static mel frequency cepstral coefficients and first order dynamic differential parameters of the static mel frequency cepstral coefficients.
5. The method for detecting diastolic dysfunction based on heart sound signals according to claim 4, wherein the extracting mel-frequency cepstrum coefficients of the heart sound signals to be detected specifically comprises:
performing fast Fourier transform on the heart sound signal to be detected to obtain a frequency spectrum sequence corresponding to the heart sound signal to be detected;
performing Mel scale filtering on the frequency spectrum sequence, and calculating the logarithmic energy of the heart sound signal to be detected according to the frequency spectrum sequence after Mel scale filtering;
determining a static Mel frequency cepstrum coefficient corresponding to the heart sound signal to be detected according to the logarithmic energy;
performing first-order difference on the static Mel frequency cepstrum coefficient to obtain a first-order dynamic difference parameter corresponding to the heart sound signal to be detected;
and obtaining the Mel frequency cepstrum coefficient of the heart sound signal to be detected according to the static Mel frequency cepstrum coefficient and the first-order dynamic difference parameter.
6. The method for detecting diastolic dysfunction based on heart sound signals according to claim 5, wherein before performing fast fourier transform on the heart sound signals to be detected to obtain a spectrum sequence corresponding to the heart sound signals to be detected, the method further comprises:
filtering the heart sound signal to be detected, and framing the filtered information of the signal to be detected to obtain a heart sound frame sequence corresponding to the heart sound signal to be detected;
and carrying out windowing on each heart sound frame in the frame sequence, and taking the heart sound frame sequence subjected to windowing as a heart sound signal to be detected.
7. The method for detecting diastolic dysfunction based on heart sound signals according to claim 6, wherein the number of corresponding heart sound sampling points of any two heart sound frames in the heart sound frame sequence is the same, and overlapping exists between any two adjacent heart sound frames, and the number of overlapping heart sound sampling points is the same.
8. A computer-readable storage medium storing one or more programs executable by one or more processors to perform the steps in the method of detecting diastolic dysfunction based on heart sound signals as claimed in any of claims 1 to 7.
9. A terminal device, comprising: a processor, a memory, and a communication bus, the memory having stored thereon a computer readable program executable by the processor;
the communication bus realizes connection communication between the processor and the memory;
the processor, when executing the computer readable program, implements the steps of the method for detecting diastolic dysfunction based on heart sound signals as claimed in any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010097640.4A CN111370120B (en) | 2020-02-17 | 2020-02-17 | Heart diastole dysfunction detection method based on heart sound signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010097640.4A CN111370120B (en) | 2020-02-17 | 2020-02-17 | Heart diastole dysfunction detection method based on heart sound signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111370120A CN111370120A (en) | 2020-07-03 |
| CN111370120B true CN111370120B (en) | 2023-07-21 |
Family
ID=71212510
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010097640.4A Active CN111370120B (en) | 2020-02-17 | 2020-02-17 | Heart diastole dysfunction detection method based on heart sound signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111370120B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111863035A (en) * | 2020-07-21 | 2020-10-30 | 上海宜硕网络科技有限公司 | Method, system and equipment for recognizing heart sound data |
| CN112233798B (en) * | 2020-12-16 | 2021-03-19 | 杭州智策略科技有限公司 | Interpretable disease risk analysis system based on pathological mode and attention mechanism |
| CN116759102B (en) * | 2023-08-17 | 2023-12-05 | 北京慧兰医疗科技有限公司 | Analysis management system based on heart rehabilitation data |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030018471A1 (en) * | 1999-10-26 | 2003-01-23 | Yan Ming Cheng | Mel-frequency domain based audible noise filter and method |
| AU2001243273A1 (en) * | 2000-02-23 | 2001-09-03 | The Johns-Hopkins University | System and method for diagnosing pathologic heart conditions |
| US20050222515A1 (en) * | 2004-02-23 | 2005-10-06 | Biosignetics Corporation | Cardiovascular sound signature: method, process and format |
-
2020
- 2020-02-17 CN CN202010097640.4A patent/CN111370120B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| 《心肺音频信号的分类和应用》;陈少琼;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;20170215;正文2-4章 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111370120A (en) | 2020-07-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111370120B (en) | Heart diastole dysfunction detection method based on heart sound signals | |
| US6135966A (en) | Method and apparatus for non-invasive diagnosis of cardiovascular and related disorders | |
| Karhade et al. | Time–frequency-domain deep learning framework for the automated detection of heart valve disorders using PCG signals | |
| US20220093215A1 (en) | Discovering genomes to use in machine learning techniques | |
| Guo et al. | Analysis and recognition of traditional Chinese medicine pulse based on the hilbert-huang transform and random forest in patients with coronary heart disease | |
| CN104473660B (en) | A kind of abnormal recognition of heart sound method based on sub-belt energy envelope autocorrelation characteristic | |
| CN109948396B (en) | Heart beat classification method, heart beat classification device and electronic equipment | |
| CN113436726B (en) | Automatic lung pathological sound analysis method based on multi-task classification | |
| Argha et al. | Artificial intelligence based blood pressure estimation from auscultatory and oscillometric waveforms: a methodological review | |
| Kaddoura et al. | Acoustic diagnosis of pulmonary hypertension: automated speech-recognition-inspired classification algorithm outperforms physicians | |
| KR20170064960A (en) | Disease diagnosis apparatus and method using a wave signal | |
| Maity et al. | Transfer learning based heart valve disease classification from Phonocardiogram signal | |
| CN113116321A (en) | Non-invasive continuous blood pressure measuring system based on PSO-GRNN neural network | |
| Wołk et al. | Early and remote detection of possible heartbeat problems with convolutional neural networks and multipart interactive training | |
| Shokouhmand et al. | Diagnosis of peripheral artery disease using backflow abnormalities in proximal recordings of accelerometer contact microphone (ACM) | |
| Azam et al. | Heart sound classification considering additive noise and convolutional distortion | |
| Megalmani et al. | Unsegmented heart sound classification using hybrid CNN-LSTM neural networks | |
| Fakhry et al. | Comparison of window shapes and lengths in short-time feature extraction for classification of heart sound signals | |
| Sabouri et al. | Effective features in the diagnosis of cardiovascular diseases through phonocardiogram | |
| Naveen et al. | Deep learning based classification of heart diseases from heart sounds | |
| Liu et al. | Energy-efficient intelligent pulmonary auscultation for post COVID-19 era wearable monitoring enabled by two-stage hybrid neural network | |
| Mukherjee et al. | A Visual Domain Transfer Learning Approach for Heartbeat Sound Classification | |
| de Lima Hedayioglu et al. | A Survey of Audio Processing Algorithms for Digital Stethoscopes. | |
| Gupta et al. | Heart Disease Prediction using Machine learning algorithms | |
| El-Dahshan et al. | An Efficient Computational Approach for Phonocardiogram Signals Analysis and Normal/Abnormal heart sounds diagnosis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |