An electrocardiosignal quality detection method, device and computer readable medium

Abstract

The application provides an electrocardiosignal quality detection method and device and a computer readable medium. The method pre-establishes a classification model for classifying electrocardiosignal quality based on multi-dimensional features of the electrocardiosignal, and optimizes the features of the established model, adjusts the number of hidden neurons, and trains the model based on the optimization and adjustment, to finally construct a signal quality detection model based on signal quality classification, which is used for quality detection of the electrocardiosignal. Thus, the application can realize automatic quality detection of the electrocardiosignal without relying on manual detection methods such as visual inspection, and the model performance is improved through feature selection optimization and hidden neuron number adjustment, which accordingly improves the electrocardiosignal quality detection speed and accuracy based on the model.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of intelligent medicine, and particularly relates to an electrocardiosignal quality evaluation method and device and a computer readable medium. BACKGROUND

[0002] An electrocardiogram (ECG) represents the periodic activity of the heart in the form of electrical signals, and is the most commonly used and most important examination for determining cardiovascular diseases in clinical practice. A dynamic electrocardiogram obtained by continuously monitoring 24-hour electrocardiosignals can provide accurate and reliable basis for diagnosing various cardiovascular diseases.

[0003] With the development of health monitoring devices, in addition to the use of Holter electrocardiosignal monitoring devices in hospitals to capture or monitor electrocardiosignals of patients, a large number of wearable electrocardiosignal monitoring devices in different forms are rapidly spreading and used in the market. However, long-term wearable ECG monitoring is at the expense of signal quality, and the local structure and features of the ECG signal are usually destroyed and covered by different noises (baseline drift, power frequency interference, sudden change and motion artifact, etc.). In various computer-aided diagnosis (CAD) systems developed by researchers, if the untreated ECG signal is used as the input of the automatic diagnosis algorithm, the original data may be distorted and deformed, and if the noise of the ECG signal is suppressed, the local waveform morphology of the signal with good quality may also be changed, which seriously affects the feature extraction and classification recognition of the signal, and increases the possibility of incorrect medical conclusions. Quality evaluation of the ECG signal plays an important role in health monitoring and reliable diagnosis of cardiovascular diseases.

[0004] Current ECG signal quality detection is usually performed by visual inspection to identify and delete damaged segments of the electrocardiosignal. However, manual quality control is not scalable for long recordings and is inefficient, and therefore an automatic implementation is needed for long-term electrocardiogram quality detection. SUMMARY

[0005] Therefore, the present application provides an electrocardiosignal quality detection method, device and computer readable medium, which can accurately, quickly and efficiently perform automatic quality detection on the electrocardiosignal to remove ECG signals with unacceptable quality.

[0006] The specific scheme is as follows:

[0007] An electrocardiosignal quality detection method comprises the following steps:

[0008] acquiring a target electrocardiosignal to be processed;

[0009] performing feature extraction on the target electrocardiosignal to obtain corresponding target signal features;

[0010] The signal quality detection model is constructed in advance, and quality detection processing is performed on the target electrocardio signal based on the target signal characteristics to obtain a signal quality detection result of the target electrocardio signal.

[0011] The signal quality detection model is a classification model for classifying electrocardio signal quality based on multi-dimensional characteristics of the electrocardio signal, and is obtained by feature selection optimization and hidden neuron number adjustment of the constructed model and model training based thereon.

[0012] Optionally, before the target electrocardio signal to be processed is acquired, the signal quality detection model is constructed.

[0013] The process of constructing the signal quality detection model includes:

[0014] The extreme learning machine is used to establish a classification model for classifying electrocardio signal quality based on multi-dimensional characteristics of the electrocardio signal.

[0015] The particle position in the particle swarm optimization algorithm is moved and the optimal particle is searched, and the particle position in the particle swarm optimization algorithm is guided by using the whale optimization algorithm to optimize the feature selection and adjust the number of hidden neurons of the extreme learning machine.

[0016] The classification model is trained based on the results of feature selection optimization and hidden neuron number adjustment.

[0017] Each position of each particle in the particle swarm in the search space represents a state of feature selection and hidden neuron number.

[0018] Optionally, the extreme learning machine is used to establish a classification model for classifying electrocardio signal quality based on multi-dimensional characteristics of the electrocardio signal, including:

[0019] A plurality of electrocardio signal samples for model training are acquired.

[0020] The sample characteristics of each electrocardio signal sample are extracted to obtain multi-dimensional characteristics of the electrocardio signal sample.

[0021] The extreme learning machine is used to establish a classification model corresponding to the multi-dimensional characteristics of the electrocardio signal sample.

[0022] Optionally, the plurality of electrocardio signal samples for model training are acquired, including:

[0023] The original electrocardio signal in the electrocardiogram is resampled, segmented, filtered and normalized to obtain a plurality of filtered and normalized electrocardio signal segments as a plurality of electrocardio signal samples.

[0024] Optionally, the sample features of each electrocardio signal sample are extracted to obtain multi-dimension features of the electrocardio signal sample, including:

[0025] The time domain features of multiple preset types and the frequency domain features of multiple preset types corresponding to each electrocardio signal sample are extracted to obtain multi-dimension features corresponding to each electrocardio signal sample respectively.

[0026] Optionally, the extreme learning machine is optimized for feature selection and adjusted for hidden neuron quantity by using the particle swarm optimization algorithm to move the positions of particles in the particle swarm and search for optimal particles, and by using the whale optimization algorithm to intervene in the search for optimal particles and guide the positions of particles in the particle swarm optimization algorithm, including:

[0027] Step one, initialize a particle set of a preset scale and the speed corresponding to each particle in the set;

[0028] Step two, based on the multi-dimension features of the electrocardio signal sample, train the classification model corresponding to each particle in the particle set at the current position, and determine the fitness of each particle in the particle set at the current position based on the training result;

[0029] Step three, for each particle in the particle swarm, select the position corresponding to the best fitness of the particle in the search process to obtain a best particle combination of the particle swarm composed of particles with the best fitness, and determine a first target particle with the best fitness in the particle swarm at the current position; determine the next position of each particle in the particle swarm based on the best particle combination and the first target particle;

[0030] Step four, adjust the next position of each particle in the particle swarm using the whale optimization algorithm; take the adjusted next position of each particle in the particle swarm as a new current position, and go to step two to iterate the processing process combining the particle swarm optimization algorithm and the whale optimization algorithm until the iteration process is ended when the end condition is met;

[0031] Step five, determine a second target particle with the best fitness in the best particle combination obtained when the iteration is ended, and take the feature selection result and the number of hidden neurons represented by the second target particle as the result of feature selection optimization and hidden neuron quantity adjustment.

[0032] Optionally, the classification model is trained based on the result of feature selection optimization and hidden neuron quantity adjustment, including:

[0033] The classification model is trained based on the feature selection and the number of hidden neurons represented by the target particle through the multi-dimension features of the electrocardio signal sample.

[0034] An electrocardio signal quality detection device, comprising:

[0035] a data acquisition unit configured to acquire a target electrocardio signal to be processed;

[0036] a feature extraction unit configured to perform feature extraction on the target electrocardio signal to obtain a corresponding target signal feature;

[0037] a quality detection unit configured to perform quality detection processing on the target electrocardio signal based on the target signal feature by using a pre-constructed signal quality detection model to obtain a signal quality detection result of the target electrocardio signal;

[0038] The signal quality detection model is a classification model for classifying electrocardio signals based on multi-dimensional features of the electrocardio signals, and is obtained by performing feature selection optimization and hidden neuron number adjustment on the built model and model training based thereon.

[0039] A computer readable medium having stored thereon a computer program, the computer program comprising program code for performing the method of any of the preceding items.

[0040] A computer program product comprising a computer program carried on a non-transitory computer readable medium, the computer program comprising program code for performing the method of any of the preceding items.

[0041] In summary, the electrocardio signal quality detection method, device and computer readable medium provided by the present application pre-establish a classification model for classifying electrocardio signals based on multi-dimensional features of the electrocardio signals, and perform feature selection optimization and hidden neuron number adjustment on the built model and model training based thereon to finally construct a signal quality detection model based on signal quality classification for quality detection of electrocardio signals. Thus, the present application can realize automatic quality detection of electrocardio signals without relying on manual detection methods such as visual inspection, and the model performance is improved by feature selection optimization and hidden neuron number adjustment, which accordingly improves the speed and accuracy of electrocardio signal quality detection based on the model. BRIEF DESCRIPTION OF DRAWINGS

[0042] The above and other features, advantages and aspects of embodiments of the present application will become more apparent by describing in detail preferred embodiments thereof with reference to the attached drawings in which:

[0043] Figure 1 is the electrocardio signal quality detection principle diagram provided by the present application;

[0044] Figure 2 is the construction process flowchart of the signal quality detection model provided by the present application;

[0045] Figure 3 is a sliding window-based ECG signal segmentation process diagram provided by the present application;

[0046] Figure 4 is a detailed construction process diagram of a signal quality detection model provided by the present application;

[0047] Figure 5 is a flowchart of an electrocardiosignal quality detection method provided by the present application;

[0048] Figure 6 is a component structure diagram of an electrocardiosignal quality detection device provided by the present application;

[0049] Figure 7 is another component structure diagram of an electrocardiosignal quality detection device provided by the present application. DETAILED DESCRIPTION

[0050] Embodiments of the present application will be described in more detail by referring to the attached drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein, but rather, these embodiments are provided to more thoroughly and completely understand the present application. It should be understood that the drawings and embodiments of the present application are for exemplary purposes only, and are not intended to limit the scope of protection of the present application.

[0051] The term "comprising" and variations thereof as used herein are open-ended, that is "including, but not limited to". The term "based on" is "based, at least in part, on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Related definitions are given throughout the description.

[0052] It should be noted that the concepts of "first", "second", etc. mentioned in the present application are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0053] It should be noted that the modification of "one" or "multiple" mentioned in the present application is illustrative rather than limiting, and those skilled in the art should understand that, unless otherwise explicitly indicated in the context, it should be understood as "one or more".

[0054] The present application discloses an electrocardiosignal quality detection method, device, computer readable medium and computer program product, which is used to realize accurate, fast and efficient automatic quality detection of electrocardiosignals based on a quality evaluation algorithm.

[0055] The quality detection process based on the quality evaluation algorithm consists of three main stages: a pre-processing stage for filtering noise using different types of filters, a feature extraction stage, and a classification stage for classifying the electrocardiogram signal quality into two or more categories based on machine learning.

[0056] Support vector machine (SVM) is a representative technique of traditional machine learning algorithms, however, as the data set increases, its computational complexity and computation time cost are increasingly high, and currently selecting appropriate kernel and parameters is still in the research stage, and cannot well solve large problems, neural network method gradually becomes a new research direction, although the neural network method has flexibility and parallel computing advantages, but the applicant finds that, since the free parameters of the network are learned using the slow gradient descent algorithm, there are problems such as lack of efficient and fast learning algorithm and selection of hidden layer neurons.

[0057] To solve the above problems, the present application combines heuristic rules and machine learning methods based on feedforward neural networks to propose a fast and effective electrocardiogram signal quality detection method, and specifically by selecting a fast and efficient learning method and combining feature selection optimization and hidden layer neuron number adjustment to improve the classification speed and accuracy of the model, and thus improve the speed and accuracy of electrocardiogram signal quality detection.

[0058] Optionally, referring to the signal quality detection principle diagram shown in Figure 1 The basic technical idea of the present application for implementing electrocardiogram signal detection is: starting from the multiple dimension features of electrocardiogram signals, using extreme learning machine (EML) to establish a classification model, and using particle swarm optimization (PSO) algorithm and whale optimization (WOA) algorithm to optimize the feature selection and adjust the number of hidden neurons of the model, and on this basis, completing model training to obtain a signal quality detection model, and based on the signal quality detection model, dividing the ECG signal into two categories of "acceptable" signal and "unacceptable" signal (it can also be more than two classification categories, without limitation), to realize quality detection of electrocardiogram signals.

[0059] The electrocardiogram signal quality detection method of the present application needs to be based on a pre-constructed signal quality detection model, and correspondingly, the signal quality detection model is: a classification model for classifying electrocardiogram signal quality based on multiple dimension features of electrocardiogram signals, and a model obtained by feature selection optimization and hidden neuron number adjustment of the established model and model training based thereon.

[0060] Referring to Figure 2 The construction process of the signal quality detection model can be implemented as:

[0061] Step 201, using an extreme learning machine, a classification model for classifying electrocardiogram signal quality based on multi-dimensional features of the electrocardiogram signal is established.

[0062] A plurality of electrocardiogram signal samples for model training can be obtained first, then the sample features of each electrocardiogram signal sample are extracted to obtain the multi-dimensional features of the electrocardiogram signal sample, and an extreme learning machine is used to establish a classification model corresponding to the multi-dimensional features of the electrocardiogram signal sample.

[0063] Among them, the original electrocardiogram signal in the electrocardiogram for model training can be resampled, segmented, filtered and normalized to obtain a plurality of electrocardiogram signal segments that have been filtered and normalized. The plurality of electrocardiogram signal segments can be regarded as a plurality of electrocardiogram signal samples.

[0064] The process can be further implemented as:

[0065] 11) For an electrocardiogram signal x(n) with a data length of N, resample it to a sampling rate of f s .

[0066] 12) Use a sliding window with a length of t to time segment the resampled ECG signal, each segment has a duration of t and a data length of L.

[0067] The ECG signal segmentation process based on the sliding window is shown in Figure 3 .

[0068] 13) The data of each t time segment is passed through a band-pass filter to reduce the interference of high-frequency and low-frequency noise, and through a median filter to reduce the interference of baseline drift noise.

[0069] Optionally, for the data of each t time segment, a zero-phase third-order Butterworth band-pass filter with a cutoff frequency of 0.25 Hz and 40 Hz can be used, but not limited to, to reduce the interference of high-frequency and low-frequency noise, and a median filter with a sliding window length of l can be used to reduce the interference of baseline drift noise.

[0070] 14) According to the maximum and minimum normalization method convert the data to the range of [0, 1], where max(x) is the maximum value of the ECG segment and min(x) is the minimum value of the ECG segment.

[0071] In the feature extraction stage, the features of each ECG signal segment can be extracted for model training, and the extracted features include a plurality of preset types of time domain features and a plurality of preset types of frequency domain features.

[0072] The extracted plurality of preset types of time domain features include but are not limited to:

[0073] a1, skewness Symmetry of the signal, where L is the length of the data, μ is the estimated mean, and σ is the estimated standard deviation of the data distribution.

[0074] a2, Kurtosis An indicator of outliers and irregularities in the ECG segment, where L is the length of the data, μ is the estimated mean, and σ is the estimated standard deviation of the data distribution.

[0075] a3, Root Mean Square The higher the value, the noisier the signal, where L is the length of the data.

[0076] a4, RR interval variability where R peak detection is performed using the Pan-Tompkins algorithm, σ RR denotes the RR interval mean estimate, μ RR denotes the RR interval standard deviation estimate.

[0077] The extracted multiple pre-set types of frequency domain features, including but not limited to:

[0078] b1, ECG signal energy Can distinguish between high-amplitude peaks or a constant line caused by loss of contact with the sensor, where L is the length of the data.

[0079] b2, QRS complex power spectrum where the peak energy of the QRS complex is mainly concentrated in 4Hz~12Hz, and the ECG signal energy is mainly concentrated in 0.25Hz~40Hz, f s is the frequency, and P(f s ) is the power spectrum of x(n).

[0080] b3, Relative power of unremoved baseline drift noise where the baseline drift frequency is generally lower than 1Hz and is difficult to completely eliminate, f s is the frequency, and P(f s ) is the power spectrum of x(n).

[0081] Step 202, by using the particle swarm optimization algorithm to move the position of the particles in the particle swarm and search for the optimal particles, and by using the whale optimization algorithm to guide the position of the particles in the particle swarm optimization algorithm, the extreme learning machine is optimized for feature selection and the number of hidden neurons is adjusted.

[0082] where each position corresponding to each particle in the particle swarm in the search space represents a state of feature selection and the number of hidden neurons.

[0083] The process can be further implemented as:

[0084] Step one, initialize a preset scale of particle set and the speed corresponding to each particle in the set.

[0085] Optionally, the particle set with a scale of S and the speed vector corresponding to each particle can be randomly initialized. The particle is represented by H-bit binary, i.e. Where t represents the current t-th iteration, the first 7 bits are representing the selection of feature j, representing that feature j is not selected, and the last (H-7) bits represent the number of hidden neurons in the ELM classifier (classification model) after being converted from binary to decimal.

[0086] Step two, based on the multi-dimensional features of the electrocardio signal sample, the classification model corresponding to each particle in the particle set at the current position is trained respectively, and the fitness corresponding to each particle at the current position is determined based on the training result.

[0087] The process specifically includes:

[0088] 21) Randomly initialize the input weight V of the hidden neuron, V ∈ R H×n and the bias b, b ∈ R H , and the number of hidden neurons is determined by the last (H-7) bits of the particle p i,t , where n is the number of samples in the training set.

[0089] For each particle, the following processing is performed:

[0090] 22) Get the output of the hidden neuron through the activation function:

[0091]

[0092] 23) Calculate the output weight W = YY h -1 through the class label Y.

[0093] Where the class label Y is a vector, specifically: for each feature of the sample, the class component 0, 1 (unacceptable, acceptable) corresponding to the feature is labeled, and the vector is composed of the labeled class components.

[0094] 24) Train the ELM classification model, use the trained (V, b, W) to classify the test sample, and calculate the accuracy acc i of the classification = number of correctly classified samples / total number of samples.

[0095] 25) Calculate the fitness corresponding to the particle where w A is the corresponding weight of acc i,t , and p i,tFor vectors (i.e., particle vectors), For vector p i,t The single element value in w F The weights for the number of selected features.

[0096] Step 3: For each particle in the particle swarm, select the position with the best fitness during the search process to obtain the best particle combination of the particle swarm consisting of particles in each best fitness position, and determine the first target particle with the best fitness in the particle swarm at the current position; determine the next position of each particle in the particle swarm based on the best particle combination and the first target particle.

[0097] Optimal fitness means the highest fitness.

[0098] Specifically, the fitness can be calculated for each particle in each iteration. i,t To update the optimal position p of each particle. bi,t (p in the first iteration) bi,t =p i,t Based on this update process, for each particle in the particle swarm, the position with the best applicability corresponding to that particle during the search process is selected, and the optimal particle combination of the particle swarm consisting of particles in each position with the best applicability is obtained.

[0099] After obtaining the current optimal particle combination and the first target particle, the velocity of each particle is first updated based on the optimal particle combination and the first target particle. Let p be the first target particle with the best fitness in the particle swarm at the current position. g,t Then the velocity of each particle in the particle swarm can be updated as follows:

[0100] v i,t+1 =wv i,t +c1r1(p bi,t -p i,t )+c2r2(p g,t -p i,t );

[0101] Where w is the inertia weighting factor, r and r2 are random variables in the range [0,1], and c1 and c2 are acceleration constants, which are used to adjust the relative velocity with respect to the optimal particle and the global position, respectively.

[0102] Then, the position of each particle in the next iteration is calculated using the updated particle velocities; that is, the next position p of each particle in the particle swarm. i,t+1 =p i,t +v i,t .

[0103] Step four, adjust the next position of each particle in the particle swarm using the whale optimization algorithm; take the adjusted next position of each particle in the particle swarm as a new current position, and go to step two to iterate the processing procedure combining the particle swarm optimization algorithm and the whale optimization algorithm until the end condition is met to end the iteration procedure.

[0104] The end condition can be, but is not limited to, that the number of iterations reaches a set number.

[0105] In this step, the process of adjusting the next position of each particle in the particle swarm using the whale optimization algorithm can be implemented as follows:

[0106] 31) define the maximum number of iterations z max , z represents the current iteration number, z = 1, 2, …, z max .

[0107] 32) randomly generate a probability pro ∈ [0, 1], and select an update method according to the size of pro:

[0108] a, pro ≥ 0.5, p i,t (z+1) = |p bi,t (z) - p i,t (tz) | · e r·l · cos(2πl) · p bi,t (z), where r is a constant defining the shape of the logarithmic spiral, and l is a uniformly distributed random vector.

[0109] b, pro < 0.5, calculate the parameter where r1 and r2 are random vectors in the range [0, 1], and a linearly decreases from 2 to 0. If |A| < 1, p bi,t (z) - A·|C·p bi,t (z) - p i,t (z) |, if |A| ≥ 1, p i,t (z+1) = p rand (z) - A·|C·p rand (z) - p i (z) |, where p rand (z) is a randomly generated combination.

[0110] 33) repeat step 32) until the maximum number of iterations z max is reached, and stop, and take the next position of each particle obtained at this time as the final adjusted next position of each particle in the particle swarm.

[0111] It should be noted that the iteration here is the iteration within step four, which is different from the iteration performed by going to step two in the above (which is essentially an iteration outside step four).

[0112] Step five, determining the second target particle with the best fitness in the best particle combination obtained at the end of the iteration (the iteration outside step four), and selecting the feature selection result and the number of hidden neurons represented by the second target particle as the result of feature selection optimization and hidden neuron number adjustment.

[0113] Step 203, training the classification model based on the result of feature selection optimization and hidden neuron number adjustment.

[0114] Among them, the classification model can be trained based on the feature selection result and the number of hidden neurons represented by the second target particle through the multi-dimensional features of the electrocardiogram signal sample.

[0115] Further, the final classification model can be trained based on the feature selection result and the number of hidden neurons represented by the second target particle according to the processing of 21)~24) in step two. The input of the classification model is the feature set of the ECG signal segment in the training set, and during the training process, the feature data in the first 7 positions of the second target particle is replaced by 0, and the number of hidden neurons is the decimal value corresponding to the (H-7) position of the second target particle.

[0116] For reference Figure 4 The detailed construction process diagram of the signal quality detection model is provided.

[0117] For reference Figure 5 As shown in the electrocardiogram signal quality detection method flowchart, based on the constructed signal quality detection model, the electrocardiogram signal quality detection method of the present application specifically includes the following processes:

[0118] Step 501, obtaining a target electrocardiogram signal to be processed.

[0119] The original electrocardiogram signal in the electrocardiogram to be quality detected can be processed through a series of processes such as resampling, segmentation, filtering and normalization to obtain each ECG signal segment to be quality detected. Each ECG signal segment to be quality detected can be regarded as a target electrocardiogram signal to be processed.

[0120] Step 502, extracting features of the target electrocardiogram signal to obtain corresponding target signal features.

[0121] Specifically, a plurality of preset types of time domain features and frequency domain features of the target electrocardiogram signal can be extracted as target signal features of the target electrocardiogram signal. The extraction process is similar to the feature extraction process of the training sample described above, and specific reference can be made to the description of the feature extraction process of the training sample above, which will not be described in detail.

[0122] In step 503, the pre-constructed signal quality detection model is used to perform quality detection processing on the target ECG signal based on the target signal features, to obtain a signal quality detection result of the target ECG signal.

[0123] Then, the target signal features corresponding to the target ECG signal, such as various types of time domain features and various types of frequency domain features corresponding to the target ECG signal, are input into the signal quality detection model. The model processes the input various features based on the feature selection result and the corresponding number of hidden neurons determined in the training phase, realizes quality classification of the target ECG signal, determines whether the quality category is "acceptable" or "unacceptable", and thus realizes quality detection of the target ECG signal.

[0124] In summary, the ECG signal quality detection method provided in the present application pre-establishes a classification model for classifying ECG signal quality based on multi-dimensional features of the ECG signal, and optimizes the feature selection and adjusts the number of hidden neurons of the established model, and trains the model based thereon, to finally construct a signal quality detection model based on signal quality classification, for quality detection of the ECG signal. Thus, the present application can realize automatic quality detection of the ECG signal without relying on manual detection methods such as visual inspection, and the performance of the model is improved through feature selection optimization and adjustment of the number of hidden neurons, which correspondingly improves the speed and accuracy of ECG signal quality detection based on the model.

[0125] Corresponding to the above-mentioned ECG signal quality detection method, the present application also provides an ECG signal quality detection device. The composition structure of the device is shown in Figure 6 and includes:

[0126] A data acquisition unit 10 is configured to acquire a target ECG signal to be processed.

[0127] A feature extraction unit 20 is configured to extract features from the target ECG signal to obtain corresponding target signal features.

[0128] A quality detection unit 30 is configured to use a pre-constructed signal quality detection model to perform quality detection processing on the target ECG signal based on the target signal features, to obtain a signal quality detection result of the target ECG signal.

[0129] The signal quality detection model is a classification model for classifying ECG signal quality based on multi-dimensional features of the ECG signal, and is obtained by optimizing feature selection and adjusting the number of hidden neurons of the established model and training the model based thereon.

[0130] In an embodiment, referring to Figure 7As shown, the apparatus can further include a model construction unit 40 configured to construct the signal quality detection model.

[0131] The model construction unit 40 constructs the signal quality detection model by performing the following steps:

[0132] The extreme learning machine is used to establish a classification model for classifying the quality of the electrocardiogram signal based on multi-dimensional features of the electrocardiogram signal.

[0133] The extreme learning machine is optimized for feature selection and the number of hidden neurons is adjusted by moving the particles in the particle swarm and searching for the optimal particles using the particle swarm optimization algorithm, and guiding the particle position in the particle swarm optimization algorithm using the whale optimization algorithm.

[0134] The classification model is trained based on the results of the feature selection optimization and the number of hidden neurons adjustment.

[0135] Each position in the particle swarm corresponds to a state of feature selection and the number of hidden neurons in the search space.

[0136] In an embodiment, when the model construction unit 40 uses the extreme learning machine to establish a classification model for classifying the quality of the electrocardiogram signal based on multi-dimensional features of the electrocardiogram signal, it is specifically configured to:

[0137] Obtain a plurality of electrocardiogram signal samples for model training.

[0138] Extract the sample features of each electrocardiogram signal sample to obtain the multi-dimensional features of the electrocardiogram signal sample.

[0139] Use the extreme learning machine to establish a classification model corresponding to the multi-dimensional features of the electrocardiogram signal sample.

[0140] In an embodiment, when the model construction unit 40 obtains a plurality of electrocardiogram signal samples for model training, it is specifically configured to:

[0141] Resample, segment, filter and normalize the original electrocardiogram signal in the electrocardiogram to obtain a plurality of filtered and normalized electrocardiogram signal segments as the plurality of electrocardiogram signal samples.

[0142] In an embodiment, when the model construction unit 40 extracts the sample features of each electrocardiogram signal sample to obtain the multi-dimensional features of the electrocardiogram signal sample, it is specifically configured to:

[0143] Extract a plurality of pre-set types of time domain features and a plurality of pre-set types of frequency domain features corresponding to each electrocardiogram signal sample respectively to obtain the multi-dimensional features corresponding to each electrocardiogram signal sample respectively.

[0144] In an embodiment, the model building unit 40 is specifically used for:

[0145] Step one, initializing a preset scale of particle set and the speed corresponding to each particle in the set;

[0146] Step two, based on the multi-dimensional features of the electrocardiosignal samples, training the classification model corresponding to each particle in the particle set at the current position, and determining the fitness corresponding to each particle in the particle set at the current position based on the training result;

[0147] Step three, for each particle in the particle group, selecting the position corresponding to the best fitness of the particle in the search process to obtain the best particle combination of the particle group composed of the particles with the best fitness, and determining the first target particle with the best fitness in the particle group at the current position; determining the next position of each particle in the particle group based on the best particle combination and the first target particle;

[0148] Step four, adjusting the next position of each particle in the particle group using the whale optimization algorithm; taking the adjusted next position of each particle in the particle group as the new current position, and turning to step two to iterate the processing process combining the particle swarm optimization algorithm and the whale optimization algorithm until the end condition is met to end the iteration process;

[0149] Step five, determining the second target particle with the best fitness in the best particle combination obtained when the iteration ends, and taking the feature selection result and the number of hidden neurons represented by the second target particle as the result of feature selection optimization and hidden neuron number adjustment.

[0150] In an embodiment, the model building unit 40 is specifically used for:

[0151] Training the classification model based on the feature selection and the number of hidden neurons represented by the target particle through the multi-dimensional features of the electrocardiosignal samples.

[0152] For the electrocardiosignal quality detection device provided by the embodiments of the present application, since it corresponds to the electrocardiosignal quality detection method provided by the method embodiment described above, the description is relatively simple, and the relevant similarities can be referred to the description of the method embodiment above, which will not be described in detail here.

[0153] The present application also provides a computer readable medium having a computer program stored thereon, the computer program comprising program codes for executing the electrocardiosignal quality detection method provided by the method embodiment described above.

[0154] In the context of this application, a computer readable medium (machine readable medium) can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine readable medium can be a machine readable signal medium or a machine readable storage medium. A machine readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine readable storage medium can include, but are not limited to, an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0155] Note that the computer readable medium described above in the present application can be a computer readable signal medium or a computer readable storage medium or any combination thereof. The computer readable storage medium, for example, can be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, the computer readable storage medium can be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. In the present application, the computer readable signal medium can include a data signal propagated in a baseband or as a carrier wave in a propagated data signal, which bears computer readable program code. Such a propagated data signal can take on many forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer readable medium can be transmitted by any suitable medium, including but not limited to wire, cable, RF (radio frequency), etc., or any suitable combination of the foregoing.

[0156] The computer readable medium described above can be contained in an electronic device; or can exist separately without being assembled into an electronic device.

[0157] The application also provides a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the electrocardiosignal quality detection method as provided by the above method embodiments.

[0158] In particular, according to embodiments of the application, the processes described above with reference to the flowcharts can be implemented as a computer software program. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, the above-mentioned functions defined in the method of the embodiments of the application are performed.

[0159] In summary, the electrocardiosignal quality detection method, device, computer-readable medium and computer program product provided by the application have at least the following technical advantages:

[0160] a) A fast and effective electrocardiosignal quality detection method is proposed by combining heuristic rules and machine learning methods based on feedforward neural networks. The model selection problem is solved by detecting the best subset of available features, avoiding many difficulties of traditional gradient-based learning algorithms, such as stopping criteria, learning rate and learning time, etc.

[0161] b) Full automation of electrocardiosignal quality detection can be achieved, and only the classification result vector needs to be modified for multi-classification problems. Compared with traditional classification techniques, it has superior generalization ability and fast learning speed, and can be used for applications that require fast prediction and response capabilities.

[0162] It should be noted that although the subject matter has been described in language specific to structural features and / or methodological acts, the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms of implementing the claims.

[0163] While the above discussion contains many specific implementation details, these should not be construed as limiting the scope of the application but merely as illustrating typical implementations of the application. Certain features that are described in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable sub-combination.

[0164] The above description is only the preferred embodiment of the present application and the explanation of the technical principles. It should be understood by those skilled in the art that the application scope of the present application is not limited to the technical solutions with the specific combination of the above technical features, and should also cover other technical solutions formed by combining the above technical features or their equivalent features without departing from the application concept. For example, the technical solutions formed by replacing the above features with the technical features with similar functions applied in the present application (but not limited to) with each other.

Claims

1. A method of electrocardiosignal quality detection, characterized in that, The method comprises the following steps: acquiring a target electrocardio signal to be processed; extracting features of the target electrocardio signal to obtain corresponding target signal features; performing quality detection processing on the target electrocardio signal based on the target signal features by using a pre-constructed signal quality detection model to obtain a signal quality detection result of the target electrocardio signal; wherein the signal quality detection model is a classification model for classifying electrocardio signal quality based on multi-dimensional features of the electrocardio signal, and the model is obtained by optimizing the features and adjusting the number of hidden neurons of the model and then training the model; before acquiring the target electrocardio signal to be processed, the method further comprises constructing the signal quality detection model; the process of constructing the signal quality detection model comprises: using an extreme learning machine to establish a classification model for classifying electrocardio signal quality based on multi-dimensional features of the electrocardio signal; optimizing the features and adjusting the number of hidden neurons of the extreme learning machine by moving and searching for the optimal particles in the particle swarm using a particle swarm optimization algorithm and guiding the particle positions in the particle swarm optimization algorithm using a whale optimization algorithm; training the classification model based on the results of the feature optimization and the hidden neuron number adjustment; wherein each position of each particle in the particle swarm in the search space represents a state of feature selection and hidden neuron number; the process of optimizing the features and adjusting the number of hidden neurons of the extreme learning machine by moving and searching for the optimal particles in the particle swarm using a particle swarm optimization algorithm and guiding the particle positions in the particle swarm optimization algorithm using a whale optimization algorithm comprises: step one, initializing a particle set of a pre-set size and the speed corresponding to each particle in the set; step two, training the classification model corresponding to each particle in the particle set at the current position based on the multi-dimensional features of the electrocardio signal sample, and determining the fitness of each particle corresponding to the classification model at the current position based on the training result; step three, for each particle in the particle swarm, selecting the position corresponding to the particle with the best fitness in the search process to obtain a best particle combination of the particle swarm composed of particles with the best fitness, and determining a first target particle with the best fitness in the particle swarm at the current position; determining the next position of each particle in the particle swarm based on the best particle combination and the first target particle; step four, adjusting the next position of each particle in the particle swarm using a whale optimization algorithm; taking the adjusted next position of each particle in the particle swarm as a new current position, and returning to step two to iterate the processing process combining the particle swarm optimization algorithm and the whale optimization algorithm until the end condition is met; step five, determining a second target particle with the best fitness in the best particle combination obtained when the iteration ends, and taking the feature selection result and the number of hidden neurons represented by the second target particle as the results of the feature optimization and the hidden neuron number adjustment.

2. The method of claim 1, wherein, The use limit learning machine, the classification model for classifying electrocardiogram signal quality based on multi-dimensional characteristics of electrocardiogram signal is established, including: Obtaining a plurality of electrocardiogram signal samples for model training; Extracting sample characteristics of each electrocardiogram signal sample to obtain multi-dimensional characteristics of electrocardiogram signal sample; Using limit learning machine, establishing classification model corresponding to multi-dimensional characteristics of electrocardiogram signal sample.

3. The method of claim 2, wherein, The method comprises the following steps: Resampling, segmenting, filtering and normalizing the original electrocardiogram signal in the electrocardiogram to obtain a plurality of filtered and normalized electrocardiogram signal segments as a plurality of electrocardiogram signal samples.

4. The method of claim 2, wherein, The method comprises the following steps: Extracting a plurality of preset types of time domain characteristics and a plurality of preset types of frequency domain characteristics corresponding to each electrocardiogram signal sample to obtain multi-dimensional characteristics corresponding to each electrocardiogram signal sample.

5. The method of claim 1, wherein, The method comprises the following steps: Based on the feature selection and the number of hidden neurons represented by the target particle, the classification model is trained through the multi-dimensional characteristics of the electrocardiogram signal sample.

6. An electrocardiosignal quality detection apparatus, characterized in that It comprises: Data acquisition unit, used for acquiring target electrocardiogram signal to be processed; Feature extraction unit, for extracting features of the target electrocardiogram signal to obtain corresponding target signal characteristics; Quality detection unit, for using a pre-constructed signal quality detection model to perform quality detection processing on the target electrocardiogram signal based on the target signal characteristics to obtain a signal quality detection result of the target electrocardiogram signal; Wherein, the signal quality detection model is: a classification model for classifying electrocardiogram signal quality based on multi-dimensional characteristics of electrocardiogram signal, and the model is obtained by feature selection optimization and hidden neuron number adjustment of the model and model training based on the model; The process of constructing the signal quality detection model comprises: Using limit learning machine, a classification model for classifying electrocardiogram signal quality based on multi-dimensional characteristics of electrocardiogram signal is established; By using particle swarm optimization algorithm to move the position of particles in particle swarm and search for optimal particles, and by using whale optimization algorithm to guide the position of particles in particle swarm optimization algorithm, the limit learning machine is optimized and adjusted in feature selection and hidden neuron number; Based on the results of feature selection optimization and hidden neuron number adjustment, the classification model is trained; Wherein, each position corresponding to each particle in the particle swarm in the search space represents a state of feature selection and hidden neuron number; The method comprises the following steps: Step one, initialize a preset size of particle set and the speed corresponding to each particle in the set; Step two, based on the multi-dimensional features of the electrocardiosignal samples, the classification model corresponding to each particle in the particle set at the current position is trained respectively, and the fitness corresponding to each particle at the current position is determined based on the training result; Step three, for each particle in the particle swarm, the position corresponding to the best fitness of the particle in the search process is selected, the best particle combination of the particle swarm composed of the particles with the best fitness is obtained, and the first target particle with the best fitness in the particle swarm at the current position is determined; the next position of each particle in the particle swarm is determined based on the best particle combination and the first target particle; Step four, the whale optimization algorithm is used to adjust the next position of each particle in the particle swarm; the adjusted next position of each particle in the particle swarm is taken as a new current position, and step two is turned to, so as to iterate the processing process combining the particle swarm optimization algorithm and the whale optimization algorithm until the end condition is met, and the iteration process is ended; Step five, the second target particle with the best fitness in the best particle combination obtained when the iteration is ended is determined, and the feature selection result represented by the second target particle and the number of hidden neurons are taken as the results of feature selection optimization and hidden neuron number adjustment.

7. A computer readable medium characterized by There is a computer program stored thereon, and the computer program comprises program codes for executing the method according to any one of claims 1-5.

8. A computer program product, characterised in that, It comprises a computer program carried on a non-transitory computer readable medium, and the computer program comprises program codes for executing the method according to any one of claims 1-5.

Images