CN110537910A - coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis - Google Patents
coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis Download PDFInfo
- Publication number
- CN110537910A CN110537910A CN201910878850.4A CN201910878850A CN110537910A CN 110537910 A CN110537910 A CN 110537910A CN 201910878850 A CN201910878850 A CN 201910878850A CN 110537910 A CN110537910 A CN 110537910A
- Authority
- CN
- China
- Prior art keywords
- energy
- sequence
- segment
- heart sound
- reflecting
- 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.)
- Granted
Links
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B7/00—Instruments for auscultation
- A61B7/02—Stethoscopes
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2576/00—Medical imaging apparatus involving image processing or analysis
- A61B2576/02—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part
- A61B2576/023—Medical imaging apparatus involving image processing or analysis specially adapted for a particular organ or body part for the heart
Abstract
a coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis comprises: the electrocardiosignal preprocessing module is used for filtering, denoising and normalizing the original electrocardiosignals; the heart sound signal preprocessing module is used for filtering, denoising and normalizing the original heart sound signal; the characteristic extraction and selection module is used for extracting multi-domain characteristics and characteristic selection of the processed electrocardio and heart sound signals; the five-channel cascade module is used for performing wavelet transformation and reconstruction on the cardiac sound signals and connecting the reconstructed four-frequency-band cardiac sound signals and the cardiac sound signals into a five-channel group signal; the double-input neural network classification module is composed of a multilayer perceptron, a deep learning model and an activation model, and is used for processing and analyzing electromechanical feature vectors and five-channel group signals and accurately screening the coronary heart disease in a noninvasive mode.
Description
Technical Field
the invention relates to a system for screening coronary heart disease, which is based on the combined analysis of electrocardio signals and belongs to the technical field of noninvasive screening of human diseases in biomedical engineering.
background
coronary heart disease is a major type of cardiovascular disease and one of the leading causes of death worldwide. Plaque build-up along the inner wall of the coronary arteries reduces blood flow to the myocardium, leading to the development of coronary heart disease. In severe cases, plaque rupture completely occludes the arterial lumen, causing acute myocardial infarction.
coronary angiography is currently considered the gold standard for clinical diagnosis of coronary heart disease. Unfortunately, this is an invasive technique, requires specialized surgery and considerable time and cost, and is therefore not attractive as a coronary heart disease screening method in general medical conditions.
the electrocardio and heart sound signals reflect the electrical and mechanical activity of the heart, respectively. Research shows that in resting electrocardiogram of coronary heart disease patient, there may exist T wave inversion, ST-T segment abnormality, left ventricular hypertrophy, ventricular premature beat and other symptoms; in the heart sound signal of a coronary heart disease patient, it is known that coronary artery stenosis causes an increase in high frequency components. Therefore, the electrocardio-voice signals and the heart-voice signals can be used as a promising tool for screening the coronary heart disease, and have the characteristics of no wound, high efficiency, simple and convenient operation, low cost and the like.
there have been many studies to automatically detect coronary heart disease using various electrocardiographic or heart sound databases. In studies using electrocardiosignals, most focus is on the traditional feature extraction and classification process. Regarding the research on heart sound signals, the accuracy of the current method for detecting coronary heart disease is low. In fact, it is far from sufficient to use only one of the electrocardio-or heart sound signals for detecting coronary heart disease, since some patients do not show abnormalities in the electrocardio-signals, but show up in the heart sound signals, and vice versa.
chinese patent document CN108577883A discloses a coronary heart disease screening device, which includes: the sound pick-up is used for acquiring a heart sound signal; the pulse wave sensor is used for acquiring a pulse wave signal; the electrocardio sensor is used for acquiring electrocardiosignals; … … calculating ST-segment level and QRS complex width of electrocardiosignal; combining the features extracted from the heart sound, the pulse wave and the electrocardio data with the medical history data and the basic physiological parameters of the user to form a feature vector; constructing a training sample, and constructing a coronary heart disease screening model based on a radial basis function neural network by adopting a nearest neighbor clustering algorithm; and inputting the characteristic vector into a screening model to obtain a screening result.
the electrocardio features extracted in the device only belong to a time domain level, and the information of a frequency domain and a nonlinear level is not considered; the extracted heart sound signal features only belong to information of a frequency domain level, and information in time domain, entropy and kurtosis is not considered; only the summary analysis of the characteristic level is carried out, the complete electrocardio-and heart-sound signals are not deeply learned, and the deep effective information related to the coronary heart disease is not comprehensively excavated from the whole situation to the details, so that the accuracy of the coronary heart disease screening is further improved.
Disclosure of Invention
aiming at the defects and shortcomings of the existing coronary heart disease detection method, the invention provides a non-invasive coronary heart disease screening system based on the synchronous acquisition of the electrocardio-heart sound signals and the heart sound signals, the system integrates the traditional classification method and the deep learning and carries out the joint analysis on the electrocardio-heart sound signals, so that the accuracy in the coronary heart disease screening is higher.
The invention relates to a coronary heart disease noninvasive screening system based on synchronously acquired electrocardio-and heart sound signal joint analysis, which adopts the following technical scheme:
the system comprises an electrocardiosignal preprocessing module, a heart sound signal preprocessing module, a feature extraction and selection module, a five-channel signal cascade module and a double-input neural network classification module; the characteristic extraction and selection module is respectively connected with the electrocardiosignal preprocessing module and the heart sound signal preprocessing module; the five-channel signal cascade module is respectively connected with the electrocardiosignal preprocessing module and the heart sound signal preprocessing module; the dual-input neural network classification module is respectively connected with the feature extraction and selection module and the five-channel signal cascade module;
the electrocardiosignal preprocessing module is used for performing band-pass filtering processing on the acquired electrocardiosignals, removing baseline drift and power frequency interference, and performing normalization processing after AD conversion;
the heart sound signal preprocessing module is used for carrying out high-pass filtering processing on the synchronously acquired heart sound signals, removing power frequency interference, and carrying out normalization processing after AD conversion;
the characteristic extraction and selection module is used for receiving the electrocardiosignals and the heart sound signals after normalization processing, extracting each domain characteristic of the electrocardiosignals by a time domain, frequency domain and time-frequency domain analysis method, extracting each domain characteristic of the heart sound signals by a time domain, frequency domain, energy, entropy and peak state analysis method, fusing each domain characteristic of the electrocardiosignals and the heart sound signals, screening out important characteristics by an information gain rate method and forming an electromechanical characteristic vector;
the five-channel signal cascade module receives the electrocardiosignals and the heart sound signals after the normalization processing, decomposes the heart sound signals into four-frequency-band signals through wavelet transformation and reconstruction, and combines the four-frequency-band signals and the electrocardiosignals into five-channel group signals;
the double-input neural network classification module comprises a multilayer perceptron, a deep learning model and an activation model, and the multilayer perceptron processes electromechanical feature vectors output by the feature extraction and selection module to obtain feature output; the deep learning model processes the output of the five-channel signal cascade module to obtain signal output; and combining the characteristic output and the signal output by the activation model, and carrying out nonlinear processing on the combined output by using the activation function to obtain a final coronary heart disease screening result.
the electrocardiosignal features extracted by the feature extraction and selection module comprise time and amplitude information of R waves, P waves, Q points, S points and T waves, and 62 time domain features are calculated according to the information; performing spectrum analysis on the electrocardiosignal by using discrete Fourier transform, and calculating 5 frequency domain characteristics; and performing time-frequency domain analysis on the electrocardiosignals by using a wavelet packet method, and calculating 14 time-frequency domain characteristics.
The heart sound signal features extracted by the feature extraction and selection module comprise time and amplitude information of an S1 segment, a systolic segment, an S2 segment and a diastolic segment, and 20 time domain features are calculated according to the information; respectively carrying out spectrum analysis on the 4 segments of heart sound signals by utilizing discrete Fourier transform, and calculating 98 frequency domain characteristics; meanwhile, 20 energy features, 8 entropy features and 8 kurtosis features are calculated.
the multilayer perceptron is composed of two fully connected layers of 32 and 64 neurons. It can also be said that there are two hidden layers inside the multi-layer perceptron, with 32 and 64 neurons respectively.
the deep learning model is composed of a convolutional neural network and a bidirectional gating circulation unit.
The activation model is constituted by an activation function Sigmoid.
the coronary heart disease screening process by the system comprises the following steps:
(1) the electrocardiosignal preprocessing module acquires an original electrocardiosignal, removes baseline drift and power frequency interference of the electrocardiosignal by using band-pass filtering, performs AD conversion on the electrocardiosignal, and then performs normalization processing on the converted signal by using a Z-score method; synchronously, a heart sound signal preprocessing module acquires a heart sound signal, removes power frequency interference of the heart sound signal by using high-pass filtering, performs AD conversion on the heart sound signal, and then performs normalization processing on the converted signal by using a Z-score method;
(2) The normalized electrocardiosignal is transmitted into a feature extraction and selection module for processing: respectively extracting time and amplitude information of R wave, P wave, Q point, S point and T wave of the electrocardiosignal, and calculating 62 time domain characteristics according to the information; performing spectrum analysis on the electrocardiosignal by using discrete Fourier transform, and calculating 5 frequency domain characteristics; performing time-frequency domain analysis on the electrocardiosignals by using a wavelet packet method, and calculating 14 time-frequency domain characteristics;
synchronously, the normalized heart sound signal transmission characteristic extraction and selection module carries out processing: respectively extracting time information and amplitude information of S1, S2, a systolic period section and a diastolic period section of the heart sound signal, and calculating 20 time domain characteristics according to the information; respectively carrying out spectrum analysis on the 4 segments of heart sound signals by utilizing discrete Fourier transform, and calculating 98 frequency domain characteristics; meanwhile, calculating 20 energy features, 8 entropy features and 8 kurtosis features;
(4) the feature extraction and selection module fuses the electrocardiosignal features and the heart sound signal features obtained by calculation, and performs dimension reduction screening on the features by using an information gain rate method to form electromechanical feature vectors;
(5) the preprocessed and normalized electrocardio signals and heart sound signals are transmitted into a signal cascade module, the heart sound signals are decomposed into four-frequency-band signals by adopting a four-scale wavelet transform and reconstruction method, and the four-frequency-band signals and the transmitted electrocardio signals form a group signal containing five-channel data;
(6) The electromechanical feature vectors and the group signals are transmitted into a double-input neural network analysis module, and a multilayer perceptron in the module uses a hidden layer containing 32 neurons and a hidden layer containing 64 neurons to process the electromechanical feature vectors to obtain feature output; synchronously, the deep learning model analyzes the group signals by using a convolutional neural network and a bidirectional gating cyclic unit GRU to obtain signal output; and the activation model uses a Sigmoid activation function to carry out nonlinear processing on the characteristic output and the signal output to obtain a final coronary heart disease screening result.
The invention has the beneficial effects that:
(1) the method fully considers the embodiment of the coronary heart disease in the electrocardiosignal, and extracts the characteristics of time domain, frequency domain and multi-layer and multi-angle of the time domain and the frequency domain; the method fully considers the embodiment of the coronary heart disease on the heart sound signal, and extracts information in various aspects of time domain, frequency domain, energy, entropy and kurtosis; the characteristics of the two signals are jointly and synchronously analyzed, so that the limitation of a single signal is overcome;
(2) The extracted electrocardio and heart sound signal characteristics are screened by using an information gain rate method, so that the characteristics related to the coronary heart disease are retained to the maximum extent, and the operation efficiency is improved to the maximum extent;
(3) the extracted electrocardio and heart sound signal characteristics are subjected to global analysis by using a multilayer perceptron, the original electrocardio and heart sound signals are subjected to detail analysis by using a deep learning model, effective information is deeply mined, and the coronary heart disease is screened in an all-around and multi-angle manner from global to detail and from summary to detail, so that the accuracy is greatly improved.
Drawings
FIG. 1 is a basic flow chart of the coronary heart disease screening method based on the combined analysis of electrocardio-and heart sound signals.
fig. 2 is a basic structure diagram of the multi-layer sensor of the present invention.
fig. 3 is a basic configuration diagram of a bidirectional gated loop unit GRU in the present invention.
Detailed Description
As shown in fig. 1, the coronary heart disease screening system based on the combined analysis of the electrocardio signal and the heart sound signal of the invention comprises: the device comprises an electrocardio signal preprocessing module, a heart sound signal preprocessing module, a feature extraction and selection module, a five-channel signal cascade module and a double-input neural network classification module. The characteristic extraction and selection module is respectively connected with the electrocardiosignal preprocessing module and the heart sound signal preprocessing module. The five-channel signal cascade module is respectively connected with the electrocardiosignal preprocessing module and the heart sound signal preprocessing module. The dual-input neural network classification module is respectively connected with the feature extraction and selection module and the five-channel signal cascade module.
the electrocardiosignal preprocessing module is used for filtering, denoising and normalizing the original electrocardiosignals; the heart sound signal preprocessing module is used for filtering, denoising and normalizing the original heart sound signal; the characteristic extraction and selection module is used for extracting multi-domain characteristics and characteristic selection of the processed electrocardio and heart sound signals; the five-channel cascade module is used for wavelet transformation and reconstruction of the heart sound signals and connecting the reconstructed heart sound signals and the electrocardio signals into a five-channel signal; the dual-input neural network classification module is used for classifying the input features and the five-channel signals and giving a coronary heart disease screening result.
The coronary heart disease screening system comprises the following specific steps:
1. signal pre-processing
Firstly, synchronously acquiring original electrocardiosignals and heart sound signals of a tested person, removing baseline drift and high-frequency noise interference of the electrocardiosignals by using a Butterworth band-pass filter (1-60Hz), removing power frequency interference of the electrocardiosignals by using a 50Hz trap wave, and normalizing the filtered electrocardiosignals by using a Z-score method; removing high-frequency noise interference of the heart sound signal by using a Butterworth high-pass filter (10Hz), removing power frequency interference of the heart sound signal by using a 50Hz trap wave, and performing normalization processing on the filtered heart sound signal by using a Z-score method;
2. feature extraction and selection
2.1 feature extraction of electrocardiosignals
2.1.1 time domain feature extraction of electrocardiosignals
(1) For the processed electrocardiosignals, acquiring time and amplitude information of R waves in each cardiac cycle by using a wavelet transformation and modulus maximum method, and further acquiring time and amplitude information of P waves, Q points, S points and T waves;
(2) using the time and amplitude information, the following 14 sequences were obtained:
RR interval sequence reflecting the time difference information of adjacent R waves;
A DRR interval sequence reflecting differential information of the RR interval sequence;
A PR interval sequence reflecting the time difference information of P waves and R waves in each cardiac cycle;
An RT interval sequence reflects time difference information of R waves and T waves in each cardiac cycle;
QS interval sequence, reflecting the time difference information between the Q point and the S point in each cardiac cycle;
QT interval sequence, reflecting the time difference information of Q point and T wave in each cardiac cycle;
the R wave amplitude sequence reflects the amplitude information of the R wave in each cardiac cycle;
The PQ amplitude sequence reflects the amplitude ratio information of P waves and Q points in each cardiac cycle;
PT amplitude sequence, which reflects the amplitude ratio information of P wave and T wave in each cardiac cycle;
the TP amplitude sequence reflects the amplitude ratio information of the T wave and the P wave in each cardiac cycle;
TQ amplitude sequence, reflecting the amplitude ratio information of T wave and Q wave in each cardiac cycle;
an R signal sequence reflecting the average amplitude information of the signal in a section with the width of 0.7s and taking the R wave as the center in each cardiac cycle;
the SP signal sequence reflects the average amplitude information of signals in a section which takes an S point as a starting point and a P wave as an end point in each cardiac cycle;
the SP signal sequence reflects the information of the degree of deviation of the ST segment from the electrocardio baseline in each cardiac cycle;
(3) Calculating the maximum value, the minimum value, the average value and the standard deviation of each sequence in the 14 sequences to obtain 56 sequence characteristics; calculating the maximum value, the minimum value, the average value, the variance and the standard deviation of the whole electrocardiosignal; and calculating the heart rate by using the time information of the R wave, and finally obtaining 62 characteristics of the electrocardiosignal time domain analysis.
2.1.2 frequency domain characterization of cardiac signals
(1) performing discrete Fourier transform on the processed electrocardiosignals to obtain a frequency spectrum sequence fi (i is 1, 2,.., N), wherein fs is a sampling rate;
(2) Calculating the average value avg _ Sp and the standard deviation std _ Sp of the frequency spectrum sequence fi, and further calculating the skewness Sp _ skew and the kurtosis Sp _ kur of the frequency spectrum sequence fi; meanwhile, the entropy Sp _ entropy of fi is calculated, and finally 5 characteristics of electrocardiosignal frequency domain analysis are obtained.
wherein, the calculation formula of the index is as follows:
2.1.3 time-frequency domain characteristics of electrocardiosignals
(1) Selecting a section signal with a width of 0.3s and taking an R wave as a center in each cardiac cycle of the processed electrocardiosignals, performing 6-scale wavelet transformation on the section signal by using a db2 wavelet function, selecting an approximate coefficient sequence a4 and a detail coefficient sequence di (i is 1, 2, 6) in a4 th scale, and calculating the energy of each sequence, namely E and Ebk (k is 1, 2, 6);
wherein, the energy calculation formula is as follows:
si represents a coefficient sequence.
(2) calculating the ratio of coefficient energy in different frequency bands, wherein the formula is as follows:
(3) wd _ ratio1, Wd _ ratio2, Wd _ ratio3, Wd _ ratio4 and Wd _ ratio5 of all cardiac cycles form respective sequences, and the mean value and standard deviation of each sequence are calculated, and 10 features are calculated;
(4) performing 4-scale wavelet packet decomposition on the processed whole electrocardiosignal by using a db6 wavelet function; reconstructing each node (16 nodes in total) of the 4 th scale, and calculating the energy of the reconstructed signal, which is recorded as Ek (k ═ 1, 2.., 16);
(5) And 4 characteristics of total energy Wp _ energy, wavelet energy entropy Wp _ entropy, energy ratio Wd _ ratio1 and Wd _ ratio2 of the Ek sequence are calculated. The calculation formula is as follows:
ei, Ek, since the formula needs to be traversed twice, a different subscript is used.
(6) finally, 14 characteristics of time-frequency domain analysis of the electrocardiosignals are obtained.
2.2 feature extraction of Heart Sound signals
2.2.1 time-domain feature extraction of Heart Sound signals
(1) Dividing the processed heart sound signals into four signals of an S1 section, a systolic section, an S2 section and a diastolic section by using a high-order Shannon entropy method, and further obtaining the following 10 sequences:
a CC sequence reflecting the time length information of each cardiac cycle;
IntS1 sequence, reflecting time difference information of adjacent S1;
IntS2 sequence, reflecting time difference information of adjacent S2;
IntSys sequence, reflecting time difference information of adjacent systolic periods;
IntDia sequence reflecting time difference information of adjacent diastolic period;
Ratio _ SysCC sequence, reflecting the Ratio of the time length of the contraction period in each cardiac cycle to the time length of the whole cardiac cycle;
Ratio _ DiaCC sequence reflecting the Ratio of the time length of the diastolic period to the time length of the whole cardiac cycle in each cardiac cycle;
a Ratio _ SysDia sequence reflecting the Ratio of the systolic phase time length to the diastolic phase time length in each cardiac cycle;
An Amp _ SysS1 sequence reflecting the ratio of the systolic segment average amplitude to the S1 segment average amplitude in each cardiac cycle;
An Amp _ DiaS2 sequence reflecting the ratio of the mean amplitude of the diastolic phase to the mean amplitude of the S2 phase in each cardiac cycle;
(2) and calculating the average value and the standard deviation of each sequence in the 10 sequences to obtain 20 characteristics of the time domain analysis of the heart sound signal.
2.2.2 frequency Domain characterization of Heart Sound signals
(1) Respectively carrying out discrete Fourier transform on the heart sound signals of the S1 segment, the systolic segment, the S2 segment and the diastolic segment to obtain corresponding frequency spectrum sequences:
fS1i (i ═ 1, 2, …, N), fSysi (i ═ 1, 2, …, N), fS2i (i ═ 1, 2, …, N), fDiai (i ═ 1, 2, —, N), fS is the sampling rate;
(2) Extracting a value fS120 with the frequency of 20Hz in an S1 spectrum sequence fS1i (i is 1, 2.. multidot.N), and calculating the mean value of fS120 in all cardiac cycles, namely fS120 m; similarly, with 20Hz as a starting point, 130Hz as an end point, and 10Hz as an interval, obtaining the mean values at different frequencies in all S1 spectrum sequences, and forming a new sequence, which is denoted as fS1jm (j is 20,30, 130):
(3) Similarly, obtaining mean sequences at different frequencies in all S2 spectrum sequences, which are denoted as fS2jm (j is 20,30, …,130), mean sequences at different frequencies in all systolic phase spectrum sequences, which are denoted as fSysjm (j is 20,30, …, 300), and mean sequences at different frequencies in all diastolic phase spectrum sequences, which are denoted as fDiajrn (j is 20,30, …, 300);
(4) calculating the ratio of the energy with the frequency above 200Hz in the S1 spectrum sequence fS1i (i is 1, 2,.. multidot.N) to the total energy of the S1 spectrum, and recording the ratio as a high-frequency energy ratio HF _ S1, and calculating the mean value and standard deviation of the high-frequency energy ratios HF _ S1 of all cardiac cycles, which are respectively recorded as HF _ S1m and HF _ S1 std; similarly, calculating the mean value and the labeling difference of the energy ratio of the frequency below 50Hz in the S1 section, and respectively recording the mean value and the labeling difference as LF _ S1m and LF _ S1 std;
(5) Similarly, high-frequency energy ratio and low-frequency energy ratio indexes of the S2 segment, the systolic segment and the diastolic segment are respectively calculated and are respectively marked as HF _ S2m, HF _ S2std, LF _ S2m, LF _ S2std, HF _ Sysm, HF _ Sysstd, LF _ Sysm, LF _ Sysstd, HF _ Diam, HF _ Diastd, LF _ Diastd and LF _ Diastd;
(6) And (4) accumulating all the characteristics in the steps (2), (3), (4) and (5) to finally obtain 98 characteristics of the heart sound signal frequency domain analysis.
2.2.3 energy, entropy and kurtosis characteristics of Heart Sound signals
(1) respectively calculating the energy ratio between the S1 segment, the systolic segment, the S2 segment and the diastolic segment, wherein the energy ratio is as follows:
energy _ S1ToSys, reflecting the Energy ratio of segment S1to the systolic segment;
energy _ S1ToDia, reflecting the Energy ratio of segment S1to the diastolic segment;
energy _ S2ToSys, reflecting the Energy ratio of segment S2to the systolic segment;
Energy _ S2ToDia, reflecting the Energy ratio of segment S2to the diastolic segment;
Energy _ DiaToSys, reflecting the Energy ratio of the diastolic phase to the systolic phase;
Energy _ S1Total, reflecting the Energy ratio of segment S1to the entire cardiac cycle;
energy _ SysTotal, reflecting the Energy ratio of the systolic phase to the whole cardiac cycle;
Energy _ S2Total, reflecting the Energy ratio of segment S2to the entire cardiac cycle;
energy _ digital, reflecting the Energy ratio of the diastolic phase to the entire cardiac cycle;
Energy _ HsTotal, reflecting the Energy sum of segment S1 and S2 versus the Energy of the entire cardiac cycle;
wherein, the energy ratio formula is as follows:
si and ci represent two signals therein, M is the length of si, and N is the length of ci.
And respectively calculating the mean value and the standard deviation of the energy ratios of all cardiac cycles to finally obtain 20 energy characteristics of the heart sound signals.
(2) Calculating sample entropy of heart sound signals in a systolic period, recording the sample entropy as SE _ Sys, and calculating the mean value and standard deviation of the SE _ Sys of all cardiac cycles; similarly, calculating the mean value and standard deviation of the self-defined fuzzy entropy FE _ Sys of the heart sound signals of the systolic period, the mean value and standard deviation of the sample entropy SE _ Dia of the heart sound signals of the diastolic period and the mean value and standard deviation of the self-defined fuzzy entropy FE _ Dia of the heart sound signals of the diastolic period, and finally obtaining 8 sample entropy characteristics of the heart sound signals;
(3) calculating kurtosis of the heart sound signals of the S1 segment, recording as S1_ kur, and calculating the mean value and standard deviation of S1_ kur of all cardiac cycles; similarly, calculating the mean value and standard deviation of the kurtosis Sys _ kur of the heart sound signals in the systolic period, the mean value and standard deviation of the kurtosis S2_ kur of the heart sound signals in the S2 period and the mean value and standard deviation of the kurtosis Dia _ kur of the heart sound signals in the diastolic period, and finally obtaining 8 kurtosis characteristics of the heart sound signals;
2.3 selection of features
and screening the obtained characteristics of the electrocardio-voice signals and the heart-voice signals by using an information gain rate method to form a final electromechanical characteristic vector EMV. The specific calculation formula is as follows:
wherein pi is the probability that the data record belongs to the ith class, and Sv is the record set with the characteristic A value of v in S;
3. Signal concatenation
performing 4-scale wavelet transformation on the heart sound signal PCG after the normalization processing by using a db6 wavelet function; the detail coefficients di (i ═ 1, 2, 3, 4) of each scale are reconstructed to obtain four-channel heart sound signals after reconstruction, which are marked as PCGi (i ═ 1, 2, 3, 4), and the heart sound signals are combined with the electrocardiosignals ECG after normalization processing into five-channel group signals.
4. dual input neural network analysis
the dual-input neural network module comprises a multilayer perceptron, a deep learning model and an activation model.
(1) The multilayer perceptron is composed of two full-connection layers with 32 and 64 neurons, as shown in fig. 2, the electromechanical feature vector EMV output by the feature selection and extraction module is used as input, and nonlinear transformation is carried out on the electromechanical feature vector EMV to obtain feature output Etop. The transformation formula is as follows:
Ft＝f(w*EMV+b)，
where w represents the weight matrix and b represents the bias vector.
(2) the deep learning model is composed of a convolutional neural network CNN and a bidirectional gating circulation unit GRU, a five-channel group signal output by a five-channel signal cascade module is used as input, and the five-channel group signal is learned and trained to obtain a signal output Sgop. The convolutional neural network comprises 10 convolutional blocks, each convolutional block comprises 3 end-to-end convolutional layers and is used for automatic feature extraction. The calculation formula for each convolutional layer is as follows:
where M denotes the filter size, i denotes the size of the input signature, j denotes the size of the convolution kernel, and kij denotes the convolution kernel for the ith input and the jth output.
and inputting the characteristic sequence output by the convolutional neural network CNN into a bidirectional gating circulation unit GRU for processing. As shown in fig. 3, the structure of the bidirectional gated loop unit GRU has two gates: the reset gate rt and the update gate zt control the update of information in common. The bidirectional GRU is a combination of a forward and a backward GRU to capture more modes. At time t, the hidden state of the forward GRU is calculated as:
Wherein xt is a sequence vector, i.e. the output of the convolutional neural network CNN, which is the previous state, which is the candidate state. Similarly, the output Sgop of the bidirectional GRU is calculated as follows:
(3) and combining the characteristic output Ftop of the multilayer perceptron and the signal output Sgop of the deep learning model, and inputting the combined signals into a nonlinear activation function Sigmoid in the activation model to obtain the final screening probability of the coronary heart disease. The formula is as follows:
σ＝ωx+b
wherein xi represents a vector obtained by combining the characteristic output Ftop and the signal output Sgop, and pi is the output of a nonlinear activation function, namely the coronary heart disease screening result.
5. analysis of Experimental data
the system of the present invention used data from 195 subjects from a third-class hospital, including 135 coronary heart disease patients and 60 non-coronary heart disease patients. The data of each subject comprises electrocardiosignals and heart sound signals, the electrocardiosignals and the heart sound signals are preprocessed, feature extraction and selection are carried out according to the steps 1 and 2, the electrocardiosignals are decomposed according to the step 3 and recombined with the electrocardiosignals, and the extracted features and the recombined signals are subjected to double-input neural network analysis according to the step 4 to obtain a coronary heart disease screening result. The results are as follows:
the analysis is carried out on the independent electrocardio signals, the independent heart sound signals, the decomposed heart sound signals, the five-channel group signals, and the combination of the characteristics and the five-channel group signals, and the result shows that the accuracy, the specificity and the sensitivity obtained by carrying out the combined synchronous analysis on the characteristics extracted from the electrocardio signals and the heart sound signals and the five-channel group signals are optimal.
meanwhile, compared with the research results of the invention, the results are as follows:
the first 5 studies all use some characteristics of individual electrocardio or heart sound signals, and do not consider analyzing the whole electrocardio and heart sound signals by using a deep learning model, so the accuracy of the method is not achieved.
it should be noted that the features of the electrocardiographic signal and the electrocardiographic signal, the feature extraction method, the feature selection method, and the data analysis method described in the present invention are merely embodiments of one or more embodiments, which means that features or methods, or combinations similar or equivalent to those of the present invention are all included in the present invention. The invention is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims. Further, all the portions not described in detail are prior art.
Claims (10)
1. A coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis is characterized by comprising:
the electrocardiosignal preprocessing module is used for performing band-pass filtering processing on the acquired electrocardiosignals, removing baseline drift and power frequency interference, and performing normalization processing after AD conversion;
the heart sound signal preprocessing module is used for carrying out high-pass filtering processing on the synchronously acquired heart sound signals, removing power frequency interference, and carrying out normalization processing after AD conversion;
the characteristic extraction and selection module is used for receiving the electrocardiosignals and the heart sound signals after normalization processing, extracting each domain characteristic of the electrocardiosignals by a time domain, frequency domain and time-frequency domain analysis method, extracting each domain characteristic of the heart sound signals by a time domain, frequency domain, energy, entropy and peak state analysis method, fusing each domain characteristic of the electrocardiosignals and the heart sound signals, screening out important characteristics by an information gain rate method and forming an electromechanical characteristic vector;
the five-channel signal cascade module receives the electrocardiosignals and the heart sound signals after the normalization processing, decomposes the heart sound signals into four-frequency-band signals through wavelet transformation and reconstruction, and combines the four-frequency-band signals and the electrocardiosignals into five-channel group signals;
the double-input neural network classification module comprises a multilayer perceptron, a deep learning model and an activation model, and the multilayer perceptron processes electromechanical feature vectors output by the feature extraction and selection module to obtain feature output; the deep learning model processes the output of the five-channel signal cascade module to obtain signal output; and combining the characteristic output and the signal output by the activation model, and carrying out nonlinear processing on the combined output by using the activation function to obtain a final coronary heart disease screening result.
2. the system for non-invasive screening of coronary heart disease based on joint analysis of electrocardiographic and cardiac sound signals according to claim 1, wherein the electrocardiographic signal features extracted by the feature extraction and selection module include time and amplitude information of R-wave, P-wave, Q-point, S-point and T-wave, and 62 time domain features are calculated according to the information; performing spectrum analysis on the electrocardiosignal by using discrete Fourier transform, and calculating 5 frequency domain characteristics; and performing time-frequency domain analysis on the electrocardiosignals by using a wavelet packet method, and calculating 14 time-frequency domain characteristics.
3. the system for non-invasive screening of coronary heart disease based on joint analysis of cardiac electrical and cardiac electrical signals according to claim 2, wherein the 62 time domain features are calculated as follows:
Using the time and amplitude information, the following 14 sequences were obtained:
RR interval sequence reflecting the time difference information of adjacent R waves;
A DRR interval sequence reflecting differential information of the RR interval sequence;
a PR interval sequence reflecting the time difference information of P waves and R waves in each cardiac cycle;
An RT interval sequence reflects time difference information of R waves and T waves in each cardiac cycle;
QS interval sequence, reflecting the time difference information between the Q point and the S point in each cardiac cycle;
QT interval sequence, reflecting the time difference information of Q point and T wave in each cardiac cycle;
The R wave amplitude sequence reflects the amplitude information of the R wave in each cardiac cycle;
the PQ amplitude sequence reflects the amplitude ratio information of P waves and Q points in each cardiac cycle;
PT amplitude sequence, which reflects the amplitude ratio information of P wave and T wave in each cardiac cycle;
The TP amplitude sequence reflects the amplitude ratio information of the T wave and the P wave in each cardiac cycle;
TQ amplitude sequence, reflecting the amplitude ratio information of T wave and Q wave in each cardiac cycle;
An R signal sequence reflecting the average amplitude information of the signal in a section with the width of 0.7s and taking the R wave as the center in each cardiac cycle;
the SP signal sequence reflects the average amplitude information of signals in a section which takes an S point as a starting point and a P wave as an end point in each cardiac cycle;
the SP signal sequence reflects the information of the degree of deviation of the ST segment from the electrocardio baseline in each cardiac cycle;
Calculating the maximum value, the minimum value, the average value and the standard deviation of each sequence in the 14 sequences to obtain 56 sequence characteristics; calculating the maximum value, the minimum value, the average value, the variance and the standard deviation of the whole electrocardiosignal; and calculating the heart rate by using the time information of the R wave, and finally obtaining 62 characteristics of the electrocardiosignal time domain analysis.
4. the system of claim 2, wherein the 5 frequency domain features are calculated by the following process:
performing discrete Fourier transform on the processed electrocardiosignals to obtain a frequency spectrum sequence fi;
calculating the average value avg _ Sp and the standard deviation std _ Sp of the frequency spectrum sequence fi, and further calculating the skewness Sp _ skew and the kurtosis Sp _ kur of the frequency spectrum sequence fi; meanwhile, the entropy Sp _ entropy of fi is calculated, and finally 5 characteristics of electrocardiosignal frequency domain analysis are obtained.
5. The system of claim 2, wherein the 14 time-frequency domain features are calculated by the following steps:
selecting a section signal with a width of 0.3s and taking an R wave as a center in each cardiac cycle of the processed electrocardiosignals, performing 6-scale wavelet transformation on the section signal by using a db2 wavelet function, selecting an approximate coefficient sequence a4 and a detail coefficient sequence di (i is 1, 2, 6) in a4 th scale, and calculating the energy of each sequence, namely E and Ebk (k is 1, 2, 6);
calculating the ratio of coefficient energy in different frequency bands;
The ratio of the coefficient energies of all cardiac cycles (Wd _ ratio1, Wd _ ratio2, Wd _ ratio3, Wd _ ratio4, Wd _ ratio5) constitutes the respective sequence, and the mean and standard deviation of each sequence are calculated, for 10 features;
performing 4-scale wavelet packet decomposition on the processed whole electrocardiosignal by using a db6 wavelet function; reconstructing each node (16 nodes in total) of the 4 th scale, and calculating the energy of the reconstructed signal, which is recorded as Ek (k ═ 1, 2.., 16);
calculating the total energy Wp _ energy, the wavelet energy entropy Wp _ entropy, the energy ratio Wd _ ratio1 and Wd _ ratio2 of the Ek sequence, wherein the total energy Wp _ energy, the wavelet energy entropy Wp _ entropy, the energy ratio Wd _ ratio1 and the Wd _ ratio2 are 4 characteristics;
Finally, 14 characteristics of time-frequency domain analysis of the electrocardiosignals are obtained.
6. the system of claim 1, wherein the features of the cardiac signal extracted by the feature extraction and selection module include time and amplitude information of S1 segment, systolic segment, S2 segment and diastolic segment, and 20 time domain features are calculated according to the information; respectively carrying out spectrum analysis on the 4 segments of heart sound signals by utilizing discrete Fourier transform, and calculating 98 frequency domain characteristics; meanwhile, 20 energy features, 8 entropy features and 8 kurtosis features are calculated.
7. the system of claim 6, wherein the calculation of the 20 time-domain features is performed by:
dividing the processed heart sound signals into four signals of an S1 section, a systolic section, an S2 section and a diastolic section by using a high-order Shannon entropy method, and further obtaining the following 10 sequences:
a CC sequence reflecting the time length information of each cardiac cycle;
IntS1 sequence, reflecting time difference information of adjacent S1;
IntS2 sequence, reflecting time difference information of adjacent S2;
IntSys sequence, reflecting time difference information of adjacent systolic periods;
IntDia sequence reflecting time difference information of adjacent diastolic period;
ratio _ SysCC sequence, reflecting the Ratio of the time length of the contraction period in each cardiac cycle to the time length of the whole cardiac cycle;
ratio _ DiaCC sequence reflecting the Ratio of the time length of the diastolic period to the time length of the whole cardiac cycle in each cardiac cycle;
A Ratio _ SysDia sequence reflecting the Ratio of the systolic phase time length to the diastolic phase time length in each cardiac cycle;
An Amp _ SysS1 sequence reflecting the ratio of the systolic segment average amplitude to the S1 segment average amplitude in each cardiac cycle;
an Amp _ DiaS2 sequence reflecting the ratio of the mean amplitude of the diastolic phase to the mean amplitude of the S2 phase in each cardiac cycle;
And calculating the average value and the standard deviation of each sequence in the 10 sequences to obtain 20 characteristics of the time domain analysis of the heart sound signal.
8. the system of claim 6, wherein the 98 frequency domain features are calculated by the following steps:
(1) respectively carrying out discrete Fourier transform on the heart sound signals of the S1 segment, the systolic segment, the S2 segment and the diastolic segment to obtain corresponding frequency spectrum sequences;
(2) extracting a value fS120 with the frequency of 20Hz in the S1 frequency spectrum sequence, and calculating the mean value of fS120 in all cardiac cycles, and marking as fS120 m; similarly, taking 20Hz as a starting point, 130Hz as an end point and 10Hz as an interval, obtaining the mean values of different frequencies in all the S1 spectrum sequences, and forming a new sequence, which is denoted as fS1jm (j ═ 20, 30.., 130);
(3) Similarly, obtaining mean value sequences at different frequencies in all S2 spectrum sequences, which are denoted as fS2jm (j 20, 30.., 130), mean value sequences at different frequencies in all systolic phase spectrum sequences, which are denoted as fSysjm (j 20, 30.., 300), and mean value sequences at different frequencies in all diastolic phase spectrum sequences, which are denoted as fdijm (j 20, 30.., 300);
(4) calculating the ratio of the energy with the frequency above 200Hz in the S1 spectrum sequence to the total energy of the S1 spectrum, recording the ratio as a high-frequency energy ratio HF _ S1, and calculating the average value and standard deviation of the high-frequency energy ratios HF _ S1 of all cardiac cycles, and recording the average value and the standard deviation as HF _ S1m and HF _ S1std respectively; similarly, calculating the mean value and the labeling difference of the energy ratio of the frequency below 50Hz in the S1 section, and respectively recording the mean value and the labeling difference as LF _ S1m and LF _ S1 std;
(5) similarly, high-frequency energy ratio and low-frequency energy ratio indexes of the S2 segment, the systolic segment and the diastolic segment are respectively calculated and are respectively marked as HF _ S2m, HF _ S2std, LF _ S2m, LF _ S2std, HF _ Sysm, HF _ Sysstd, LF _ Sysm, LF _ Sysstd, HF _ Diam, HF _ Diastd, LF _ Diastd and LF _ Diastd;
(6) and (4) accumulating all the characteristics in the steps (2), (3), (4) and (5) to finally obtain 98 characteristics of the heart sound signal frequency domain analysis.
9. The system of claim 6, wherein the 20 energy features, 8 entropy features, and 8 kurtosis features are calculated by:
(1) Respectively calculating the energy ratio between the S1 segment, the systolic segment, the S2 segment and the diastolic segment, wherein the energy ratio is as follows:
energy _ S1ToSys, reflecting the Energy ratio of segment S1to the systolic segment;
energy _ S1ToDia, reflecting the Energy ratio of segment S1to the diastolic segment;
energy _ S2ToSys, reflecting the Energy ratio of segment S2to the systolic segment;
energy _ S2ToDia, reflecting the Energy ratio of segment S2to the diastolic segment;
energy _ DiaToSys, reflecting the Energy ratio of the diastolic phase to the systolic phase;
Energy _ S1Total, reflecting the Energy ratio of segment S1to the entire cardiac cycle;
energy _ SysTotal, reflecting the Energy ratio of the systolic phase to the whole cardiac cycle;
Energy _ S2Total, reflecting the Energy ratio of segment S2to the entire cardiac cycle;
energy _ digital, reflecting the Energy ratio of the diastolic phase to the entire cardiac cycle;
energy _ HsTotal, reflecting the sum of the energies of segment S1 and segment S2 versus the Energy of the entire cardiac cycle.
And respectively calculating the mean value and the standard deviation of the energy ratios in all cardiac cycles to finally obtain 20 energy characteristics of the heart sound signals.
(2) Calculating sample entropy of heart sound signals in a systolic period, recording the sample entropy as SE _ Sys, and calculating the mean value and standard deviation of the SE _ Sys of all cardiac cycles; similarly, calculating the mean value and standard deviation of the self-defined fuzzy entropy FE _ Sys of the heart sound signals of the systolic period, the mean value and standard deviation of the sample entropy SE _ Dia of the heart sound signals of the diastolic period and the mean value and standard deviation of the self-defined fuzzy entropy FE _ Dia of the heart sound signals of the diastolic period, and finally obtaining 8 sample entropy characteristics of the heart sound signals;
(3) calculating kurtosis of the heart sound signals of the S1 segment, recording as S1_ kur, and calculating the mean value and standard deviation of S1_ kur of all cardiac cycles; similarly, the mean value and the standard deviation of the kurtosis Sys _ kur of the heart sound signals in the systolic period, the mean value and the standard deviation of the kurtosis S2_ kur of the heart sound signals in the S2 period, and the mean value and the standard deviation of the kurtosis Dia _ kur of the heart sound signals in the diastolic period are calculated, and finally 8 kurtosis characteristics of the heart sound signals are obtained.
10. The system of claim 1, wherein the multi-layered sensor is composed of two fully connected layers with 32 and 64 neurons; the deep learning model consists of a convolutional neural network and a bidirectional gating circulation unit; the activation model is constituted by an activation function Sigmoid.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910878850.4A CN110537910B (en) | 2019-09-18 | 2019-09-18 | Coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910878850.4A CN110537910B (en) | 2019-09-18 | 2019-09-18 | Coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110537910A true CN110537910A (en) | 2019-12-06 |
| CN110537910B CN110537910B (en) | 2021-05-04 |
Family
ID=68713941
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910878850.4A Active CN110537910B (en) | 2019-09-18 | 2019-09-18 | Coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110537910B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111798980A (en) * | 2020-07-10 | 2020-10-20 | 哈尔滨工业大学（深圳） | Complex medical biological signal processing method and device based on deep learning network |
| CN112336369A (en) * | 2020-11-30 | 2021-02-09 | 山东大学 | Coronary heart disease risk index evaluation system of multichannel heart sound signals |
| CN112494000A (en) * | 2020-11-26 | 2021-03-16 | 河北德睿健康科技有限公司 | Coronary artery stenosis visualization quantification method and equipment based on multiple paths of heart sounds |
| CN113449636A (en) * | 2021-06-28 | 2021-09-28 | 苏州美糯爱医疗科技有限公司 | Artificial intelligence-based automatic classification method for severity of aortic stenosis |
| CN113598784A (en) * | 2021-08-25 | 2021-11-05 | 济南汇医融工科技有限公司 | Arrhythmia detection method and system |
| CN114764575A (en) * | 2022-04-11 | 2022-07-19 | 山东省人工智能研究院 | Multi-modal data classification method based on deep learning and time sequence attention mechanism |
| CN115363586A (en) * | 2022-09-08 | 2022-11-22 | 山东大学 | Psychological stress grade assessment system and method based on pulse wave signals |
| CN115553784A (en) * | 2022-09-26 | 2023-01-03 | 山东大学 | Coronary heart disease assessment method and system based on electrocardio and heart sound signal coupling analysis |
| CN115640507A (en) * | 2022-12-09 | 2023-01-24 | 南京备力医疗科技有限公司 | Abnormal data screening method based on electrocardio-heart sound joint analysis |
| CN116364117A (en) * | 2023-03-23 | 2023-06-30 | 南京市浦口医院 | Low-frequency electronic auscultation system based on sound vibration microphone |
| CN117281525A (en) * | 2023-09-19 | 2023-12-26 | 山东大学 | Coronary heart disease detection system for graphic combined analysis of electrocardio and heart sound signals |
| CN117349600A (en) * | 2023-12-06 | 2024-01-05 | 厦门理工学院 | Heart sound and heart electricity combined diagnosis method and system based on dual-mode dual input |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101176659A (en) * | 2007-12-06 | 2008-05-14 | 山东大学 | Method and apparatus for detecting systema cariovasculare functional state |
| CN105748038A (en) * | 2016-02-01 | 2016-07-13 | 山东大学 | Nondestructive testing device for coronary heart disease risk indexes |
| CN108577883A (en) * | 2018-04-03 | 2018-09-28 | 上海交通大学 | A kind of Screening for coronary artery disease device, screening system and signal characteristic extracting methods |
| CN108968933A (en) * | 2018-06-23 | 2018-12-11 | 苏强 | Cardiovascular and cerebrovascular disease detection device |
| US20190105019A1 (en) * | 2017-10-05 | 2019-04-11 | EchoNous, Inc. | System and method for fusing ultrasound with additional signals |
-
2019
- 2019-09-18 CN CN201910878850.4A patent/CN110537910B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101176659A (en) * | 2007-12-06 | 2008-05-14 | 山东大学 | Method and apparatus for detecting systema cariovasculare functional state |
| CN105748038A (en) * | 2016-02-01 | 2016-07-13 | 山东大学 | Nondestructive testing device for coronary heart disease risk indexes |
| US20190105019A1 (en) * | 2017-10-05 | 2019-04-11 | EchoNous, Inc. | System and method for fusing ultrasound with additional signals |
| CN108577883A (en) * | 2018-04-03 | 2018-09-28 | 上海交通大学 | A kind of Screening for coronary artery disease device, screening system and signal characteristic extracting methods |
| CN108968933A (en) * | 2018-06-23 | 2018-12-11 | 苏强 | Cardiovascular and cerebrovascular disease detection device |
Non-Patent Citations (1)
| Title |
|---|
| 邓德秒: ""心音与心电信号检测系统研究"", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111798980A (en) * | 2020-07-10 | 2020-10-20 | 哈尔滨工业大学（深圳） | Complex medical biological signal processing method and device based on deep learning network |
| CN112494000A (en) * | 2020-11-26 | 2021-03-16 | 河北德睿健康科技有限公司 | Coronary artery stenosis visualization quantification method and equipment based on multiple paths of heart sounds |
| CN112336369A (en) * | 2020-11-30 | 2021-02-09 | 山东大学 | Coronary heart disease risk index evaluation system of multichannel heart sound signals |
| CN112336369B (en) * | 2020-11-30 | 2021-09-10 | 山东大学 | Coronary heart disease risk index evaluation system of multichannel heart sound signals |
| CN113449636B (en) * | 2021-06-28 | 2024-03-12 | 苏州美糯爱医疗科技有限公司 | Automatic aortic valve stenosis severity classification method based on artificial intelligence |
| CN113449636A (en) * | 2021-06-28 | 2021-09-28 | 苏州美糯爱医疗科技有限公司 | Artificial intelligence-based automatic classification method for severity of aortic stenosis |
| CN113598784A (en) * | 2021-08-25 | 2021-11-05 | 济南汇医融工科技有限公司 | Arrhythmia detection method and system |
| CN113598784B (en) * | 2021-08-25 | 2024-04-09 | 济南汇医融工科技有限公司 | Arrhythmia detection method and system |
| CN114764575A (en) * | 2022-04-11 | 2022-07-19 | 山东省人工智能研究院 | Multi-modal data classification method based on deep learning and time sequence attention mechanism |
| CN114764575B (en) * | 2022-04-11 | 2023-02-28 | 山东省人工智能研究院 | Multi-modal data classification method based on deep learning and time sequence attention mechanism |
| CN115363586A (en) * | 2022-09-08 | 2022-11-22 | 山东大学 | Psychological stress grade assessment system and method based on pulse wave signals |
| CN115553784A (en) * | 2022-09-26 | 2023-01-03 | 山东大学 | Coronary heart disease assessment method and system based on electrocardio and heart sound signal coupling analysis |
| CN115553784B (en) * | 2022-09-26 | 2024-04-26 | 山东大学 | Coronary heart disease assessment method and system based on electrocardio and heart sound signal coupling analysis |
| CN115640507B (en) * | 2022-12-09 | 2024-03-15 | 镜电(南京)科技发展有限公司 | Abnormal data screening method based on electrocardiographic and heart sound combined analysis |
| CN115640507A (en) * | 2022-12-09 | 2023-01-24 | 南京备力医疗科技有限公司 | Abnormal data screening method based on electrocardio-heart sound joint analysis |
| CN116364117A (en) * | 2023-03-23 | 2023-06-30 | 南京市浦口医院 | Low-frequency electronic auscultation system based on sound vibration microphone |
| CN117281525A (en) * | 2023-09-19 | 2023-12-26 | 山东大学 | Coronary heart disease detection system for graphic combined analysis of electrocardio and heart sound signals |
| CN117281525B (en) * | 2023-09-19 | 2024-03-29 | 山东大学 | Coronary heart disease detection system for graphic combined analysis of electrocardio and heart sound signals |
| CN117349600B (en) * | 2023-12-06 | 2024-01-30 | 厦门理工学院 | Heart sound and heart electricity combined diagnosis method and system based on dual-mode dual input |
| CN117349600A (en) * | 2023-12-06 | 2024-01-05 | 厦门理工学院 | Heart sound and heart electricity combined diagnosis method and system based on dual-mode dual input |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110537910B (en) | 2021-05-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110537910B (en) | Coronary heart disease noninvasive screening system based on electrocardio and heart sound signal joint analysis | |
| US20070129639A1 (en) | Methods and analysis for cardiac ischemia detection | |
| CN109259756B (en) | ECG signal processing method based on secondary neural network of unbalanced training | |
| Vijaya et al. | Arrhythmia detection through ECG feature extraction using wavelet analysis | |
| Khandait et al. | Features extraction of ECG signal for detection of cardiac arrhythmias | |
| WO2021170043A1 (en) | Electrocardiogram data classification method and classification system, computer device, and storage medium | |
| Anbalagan et al. | Analysis of various techniques for ECG signal in healthcare, past, present, and future | |
| Wu et al. | Personalizing a generic ECG heartbeat classification for arrhythmia detection: a deep learning approach | |
| Yu et al. | Ddcnn: A deep learning model for af detection from a single-lead short ecg signal | |
| Singh et al. | Short and noisy electrocardiogram classification based on deep learning | |
| Kar et al. | A technical review on statistical feature extraction of ECG signal | |
| Begum et al. | Detection of cardiomyopathy using support vector machine and artificial neural network | |
| Ghosh et al. | Introduction of boosting algorithms in continuous non-invasive cuff-less blood pressure estimation using pulse arrival time | |
| Mafi | Signal processing methods for heart rate detection using the seismocardiogram | |
| Dogan et al. | A comprehensive review of computer-based Techniques for R-peaks/QRS complex detection in ECG signal | |
| Prasad et al. | ECG signal analysis: different approaches | |
| Gad | Feature extraction of electrocardiogram signals using discrete sinc transform | |
| Bassiouni et al. | Combination of ECG and PPG signals for smart healthcare systems: Techniques, applications, and challenges | |
| Bassiouni et al. | Combination of ECG and PPG Signals for Healthcare Applications: A Survey | |
| Rajendran et al. | Classification of heart disease from ECG signals using Machine Learning | |
| Rajesh et al. | Detection and Identification of Irregularities in Human Heart Rate | |
| Ouelli et al. | AR modeling for automatic cardiac arrhythmia diagnosis using QDF based algorithm | |
| AlHosani et al. | Automatic detection of coronary artery disease (CAD) in an ECG signal | |
| Sabbah et al. | Murine Atherosclerosis Detection Using Machine Learning Under Magnetic Resonance Imaging | |
| Das | Electrocardiogram signal analysis for heartbeat pattern classification |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210414 Address after: 250101 room B307, block F, pioneer Plaza, Qilu Software Park, high tech Zone, Jinan City, Shandong Province Applicant after: JINAN HUIYI RONGGONG TECHNOLOGY Co.,Ltd. Address before: No. 27, mountain Dana Road, Ji'nan City, Shandong, Shandong Applicant before: SHANDONG University |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |