CN114176602A - Method for simultaneously positioning positions of electrocardio P wave, QRS wave and T wave based on deep learning multi-target detection - Google Patents
Method for simultaneously positioning positions of electrocardio P wave, QRS wave and T wave based on deep learning multi-target detection Download PDFInfo
- Publication number
- CN114176602A CN114176602A CN202111662087.5A CN202111662087A CN114176602A CN 114176602 A CN114176602 A CN 114176602A CN 202111662087 A CN202111662087 A CN 202111662087A CN 114176602 A CN114176602 A CN 114176602A
- Authority
- CN
- China
- Prior art keywords
- wave
- waves
- qrs
- target detection
- deep learning
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000013135 deep learning Methods 0.000 title claims abstract description 42
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 18
- 238000004364 calculation method Methods 0.000 claims description 22
- 238000000605 extraction Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 12
- 230000001629 suppression Effects 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 5
- 230000002829 reductive effect Effects 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 2
- 230000000717 retained effect Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 description 5
- 208000024172 Cardiovascular disease Diseases 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000036961 partial effect Effects 0.000 description 2
- 230000002441 reversible effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 206010003119 arrhythmia Diseases 0.000 description 1
- 230000006793 arrhythmia Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 208000031225 myocardial ischemia Diseases 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000001575 pathological effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
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
- A61B5/24—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]
- A61B5/346—Analysis of electrocardiograms
-
- 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]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/353—Detecting P-waves
-
- 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]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/355—Detecting T-waves
-
- 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]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
- A61B5/366—Detecting abnormal QRS complex, e.g. widening
-
- 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
- A61B5/7235—Details of waveform analysis
-
- 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
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- 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
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
Abstract
The invention discloses a method for simultaneously positioning positions of electrocardio P waves, QRS waves and T waves based on deep learning multi-target detection, which solves the problem that the traditional algorithm needs to rely on the QRS waves to detect the P waves and the T waves, can directly obtain the starting and ending positions of the P waves, the QRS waves and the T waves, and compared with the scheme of the deep learning heartbeat detection disclosed at present, the invention completes the establishment of simultaneously detecting the P waves and the T waves.
Description
Technical Field
The invention relates to the technical field of artificial intelligence one-dimensional data and medical electrocardio detection, in particular to a detection method for simultaneously positioning positions of P waves, QRS waves and T waves of electrocardiosignals based on deep learning target detection.
Background
The electrocardiogram is the most simple and effective tool for diagnosing cardiovascular diseases, and various cardiovascular diseases such as arrhythmia and myocardial ischemia can be diagnosed by utilizing electrocardiosignals. The electrocardiogram heart beat mainly comprises a P wave, a QRS wave group and a T wave, the P wave, the QRS wave and the T wave in different pathological states have different characteristics, and the different characteristics reflect different cardiovascular diseases, so that the detection and identification of the P wave, the QRS wave and the T wave are particularly important. Doctors routinely review and correct the results of automated electrocardiographic analysis with the aid of electrocardiographic software. If the accuracy of the electrocardio-recognition can be improved, the workload of doctors can be greatly reduced, which is beneficial to the accurate screening of cardiovascular diseases.
At present, the traditional electrocardio P wave, QRS wave and T wave identification and positioning method is to artificially design a filter to reduce noise and enhance the electrocardiosignal according to the basic rules of the electrocardiosignal and the electrophysiological activity of the heart. The artificial features are designed through difference calculation, area calculation and the like, a proper threshold value is set through an empirical value, and P waves and T waves are detected by relying on the position of QRS waves after the QRS waves are detected. However, the heart beat of the electrocardiosignal has individual difference due to different physical conditions of each person, and the electrocardiosignal may have noise interference in the acquisition process, so that the traditional signal analysis method is difficult to consider all conditions based on the characteristics of manual design, has low self-adaptability, and reduces the accuracy of heart beat detection when the electrocardiosignal beyond the characteristics of manual design occurs.
The deep learning has the characteristics of strong autonomous learning capability and high nonlinear mapping, can learn the electrocardiosignal characteristics in a large amount of training data, has high self-adaptability, and provides possibility for designing complex and high-precision electrocardiosignal P wave, QRS wave and T wave detection and identification models. Some researches at present realize the vertex detection of the QRS wave by using an artificial intelligence technology, realize the detection of the initial and end positions of the QRS wave by using a disclosed deep learning target detection algorithm, and use a disclosed feature extraction network such as AlexNet, VGG, ResNet, Squeezenet, MobileNet and the like in an electrocardiosignal feature extraction part. Because the time window of the P wave is shorter than that of the QRS wave, after the electrocardio data is compressed by using the characteristic extraction network, the characteristics of the P wave are weakened or disappeared, and the P wave detection is not facilitated.
Disclosure of Invention
In view of the above, the first aspect of the present invention is to provide a method for simultaneously locating positions of P waves, QRS waves, and T waves of electrocardio based on deep learning multi-target detection.
The purpose of the first aspect of the invention is realized by the following technical scheme:
the method for simultaneously positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection comprises the following steps:
step S1: acquiring electrocardiosignals, and performing noise reduction processing on the electrocardiosignals;
step S2: intercepting the electrocardio data for a set time length, ensuring that an electrocardio sample data set contains different conditions of P waves, QRS waves and T waves of all the electrocardio data, and marking the starting and ending positions of the P waves, the QRS waves and the T waves of the electrocardio data set to obtain a marked data set;
step S3: carrying out prior frame processing in deep learning target detection on the intercepted and marked sample data, wherein the width of the prior frame is related to the widths of a P wave, a QRS wave and a T wave, and a plurality of prior frames are generated on the sample data; converting the positions of the prior frame, the actual marked P wave, the QRS wave and the T wave into the relation of offset according to the deep learning target detection, thereby obtaining a data set required by model training;
step S4: using deep learning target detection to identify a feature extraction network used for the heartbeat detection;
step S5: inputting the training set data prepared in the step S3 into a model according to a suggestion frame detection algorithm in a deep learning target detection algorithm, and performing regression and classification operation on the model to distinguish which parts of the electrocardiogram data with set time length are suggestion frames containing P wave, QRS wave and T wave positions;
step S6: carrying out suggested frame offset data set processing;
step S7: performing secondary processing on the suggestion frames, performing intersection comparison calculation on the positions of the suggestion frames obtained in the last step and the positions of the P waves, the QRS waves and the T waves of the real marks, taking the suggestion frames which are larger than a certain threshold value as positive samples, marking the categories of the P waves, the QRS waves and the T waves of the positive samples, taking the suggestion frames which are smaller than the certain threshold value as negative samples, and coding the suggestion frames of the positive samples and the negative samples into a coefficient relation with the offset of the real frames to be used as sample data of a classification layer;
step S8: inputting the output features acquired through the feature extraction network and the classified sample data in the step S7 into a classification layer, intercepting the extracted features through a secondarily processed suggestion box, wherein the length of the suggestion box is changed into the same size and then is input into the classification layer because the lengths of the suggestion boxes are different, and obtaining the offset of the prediction box and the corresponding category of the prediction box through the classification layer;
step S9: and finally obtaining the start and end positions and the types of the P wave, the QRS wave and the T wave through model prediction of the obtained electrocardiogram data.
Further, in step S4, the feature extraction network uses a ResNet structure;
further, in step S4, a pyramid-like network with feature layers superimposed one on top of the other is used to implement an electrocardiographic data feature extraction network, two feature layers with different sizes are constructed, it is ensured that the set prior frame can contain the actually labeled P wave, QRS wave, and T wave positions, and the captured electrocardiographic data within the set time length is input to the feature extraction network to obtain electrocardiographic data features.
Further, in step S6, according to the situation that there are multiple overlapping or overlapping suggested frame positions obtained in the previous layer, a non-maximum suppression mode is used in the deep learning target detection algorithm to clear the suggested frame, but because the suggested frame processing layer is not the final result, the threshold value of the used non-maximum suppression is large, and a part of the suggested frames are retained for further processing; .
Further, in step S9, the prediction frames obtained may overlap, and the overlapping prediction frames are eliminated by using non-maximum value suppression.
Further, in the step S1, the set time period is 8 to 20S.
Further, in step S3, under the determined sampling frequency, obtaining a plurality of preselected windows, performing iou union ratio calculation in target detection by using the preselected windows and the marked data real windows, where the iou union ratio calculation mode is a type of a method in general image target detection, reducing two-dimensional calculation to one-dimensional calculation in electrocardiographic data, obtaining an intersection ratio by iou calculation, dividing prediction windows in different intersection ratio ranges into positive samples, marking negative samples and backgrounds respectively, and calculating a preselected frame and a real frame in the positive samples to obtain an offset of each preselected window to the real window.
Further, the positive direction sample is marked as 1, the background is 0, and the negative direction sample is-1;
the invention provides a device for simultaneously locating positions of P waves, QRS waves and T waves of electrocardio based on deep learning multi-target detection, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the method when executing the computer program.
It is an object of a third aspect of the invention to provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, performs the method as previously described.
The invention has the beneficial effects that: the method solves the problem that the detection of the P wave and the T wave needs to be carried out by depending on the QRS wave in the traditional algorithm, and can directly obtain the starting and ending positions of the P wave, the QRS wave and the T wave. Compared with the scheme of deep learning heart beat detection disclosed at present, the invention completes the task of simultaneously detecting P waves and T waves. Aiming at a scheme of partial deep learning heart beat detection, the invention uses a two-stage scheme in deep learning target detection and uses a feature extraction layer similar to pyramid superposition, thereby achieving the technical effect of simultaneously detecting a small target P wave and a QRS wave and a T wave which are wider than the P wave.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objectives and other advantages of the present invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings, in which:
FIG. 1 is a schematic flow diagram of the process of the present invention;
FIG. 2 is a diagram of a model architecture employed by an embodiment of the present invention;
fig. 3 is a labeled graph of a P wave, a QRS wave, and a T wave of a single heart beat;
fig. 4 is a sample graph of 10-second ecg data after labeling.
Detailed Description
Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be understood that the preferred embodiments are illustrative of the invention only and are not limiting upon the scope of the invention.
As shown in FIG. 1, the method for simultaneously positioning the positions of P waves, QRS waves and T waves of electrocardios based on deep learning multi-target detection comprises the following steps:
step S1: electrocardiosignal collection and pretreatment
Denoising the electrocardiosignals through a band-pass filter or other modes (including but not limited to conventional signal processing modes such as power frequency interference, baseline drift, electromyographic interference and the like), obtaining electrocardio data subjected to denoising processing, and normalizing the electrocardio data;
step S2: the electrocardiosignal is sliced to be intercepted for a set time length (in the embodiment, the length is 10 seconds, other ranges can be selected in the actual use process, the set time length is suggested to be 8-20s), the fragment data form an electrocardio data set, and the electrocardio data set is ensured to contain different P waves, QRS waves, T waves and different intervals.
And marking the starting and ending positions of the P wave, the QRS wave and the T wave of the electrocardiogram data set to obtain a marked data set.
Step S3: target data required by model training constructed according to labeled data samples
According to the electrocardiograph knowledge, the time length of a P wave is generally less than 0.11s, the time length of a QRS wave is 0.11-0.25s, the time length of a T wave is about 0.25s, the P wave and the QRS wave are obtained, a standard window of the T wave is obtained by rounding the waveform time length(s) of the electrocardiograph data sampling rate (fs), and taking 500hz sampling as an example, two windows of 55 and 125 (here: 500 0.11 is 55 and 500 is 0.25 is 125) are obtained, abnormal waveforms appear in the real electrocardiograph, and the actual waveform lengths are various, so that the two windows are lengthened or shortened to obtain 6 preselection windows used by the model of 27, 55, 68, 83, 125 and 187 respectively, a priori frames are generated on the electrocardiograph data by using the 6 windows with the sizes, and the 6 windows are the sizes of the priori frames. The window can also obtain a window range with concentrated distribution by clustering the mark data.
And carrying out iou merging proportion calculation in target detection by using the preselected window and the marked data real window, wherein the merging proportion calculation mode is a method type in general image target detection, two-dimensional calculation in a picture is reduced to one-dimensional calculation in electrocardiogram data, merging proportions are obtained through iou calculation, prediction windows in different merging proportion ranges are divided into positive samples, negative samples and backgrounds, the positive samples are marked as 1, the backgrounds are 0, and the negative samples are-1. And calculating the offset of each preselected window to the real window by the preselected frame and the real frame in the forward sample.
The 6 pre-selected windows are divided into two categories, one for detecting P-waves 27, 55, 68 and the second for detecting QRS-waves and T-waves 83, 125, 187. Therefore, the positive and negative samples and the background sample obtained through iou are divided, the offset is calculated with a real window of a real mark, the two samples are stored in an array, and model prediction target data are obtained. The purpose of this is to train on different feature layers in the model, so as to obtain more accurate results.
Step S4: model building
At present, deep learning target detection is used for feature extraction networks used for detecting and identifying heart beat, and the networks which are mature in the open field, such as AlexNet, VGG, inclusion, ResNet, Squeezenet and MobileNet, are mainly used. However, the P wave cannot be detected after the electrocardiogram data is compressed, so the method uses a pyramid-like network with superposed feature layers to realize the electrocardiogram data feature extraction network, constructs two feature layers with different sizes, ensures that the set prior frame can contain the actually marked positions of the P wave, the QRS wave and the T wave, and inputs the intercepted 10-second electrocardiogram data into the feature extraction network to obtain the electrocardiogram data features.
As shown in fig. 2, in the present embodiment, a ResNet structure is used as the feature extraction network, and different from the disclosed deep learning electrocardiographic target detection algorithm, the present invention uses, as the output of the feature extraction layer, 1/4 (denoted as Stage1) and 1/2(Stage2) which have been down-sampled to the original data length in the model structure, and a network layer of the last Stage (Stage3) of ResNet in the model structure, and uses different Stage layers to facilitate the superposition of the subsequent feature layers, Stage2 adjusts the number of output channels to be equal to Stage3 by convolution of 1 ^ 1, Stage3 aligns the length with Stage2 by up-sampling, and then Stage2 and Stage3 are superposed to obtain the first target detection layer P2. The number of output channels of Stage1 is adjusted to be equal to that of P2 through 1 × 1 convolution, the length of P2 is aligned with that of Stage1 through upsampling, and P2 and Stage1 are overlapped to obtain a second target detection layer P1. The length of the P2 target detection layer obtained by the above operation is equal to 1/4 of the original data length, and is used as the target detection layer of the QRS wave and the T wave. The P1 target detection layer length is equal to 1/2 length of the original data length, and is the target detection layer of the P wave. Different feature layers are used for detecting different waveforms, the QRS wave and the T wave are longer in length, a feature layer P2 with small size is used, and the P wave is shorter in length, and a larger feature layer P1 is used. The invention has the advantages that different characteristic layers are used for detecting different characteristics of the P wave, the QRS wave and the T wave, and if the P2 characteristic layer is used for detecting the P wave, the P2 characteristic layer is compressed, the size of the characteristic layer is reduced, the P wave characteristic disappears, and the P wave detection is not facilitated.
An rpn layer is connected behind the target detection layer, and a rpn layer outputs two groups of data, namely cls _ scores, for distinguishing whether waveform data of one of P waves, QRS waves and T waves exist in the electrocardio number of a preselected window, and a recommendation frame cls _ bboxes of the P waves, the QRS waves and the T waves exists in the other group. In the layer model calculation, the judgment is made by the array marked in the step (three (3)), and if the first group of data in the array detects P waves, the calculation is carried out in the corresponding P1 target detection layer. Similarly, the detection of QRS wave and T wave is calculated in the corresponding P2 target detection layer.
And S5, inputting the training set data prepared in the step S3 into a model according to a suggestion frame detection algorithm in the deep learning target detection algorithm, and distinguishing which parts of the electrocardiogram data with set time length are suggestion frames containing P wave, QRS wave and T wave positions through regression and classification operation of the model.
The method is mainly used for distinguishing which preselection frames in the electrocardiogram data comprise positions of P waves, QRS waves and T waves, and which preselection frames do not comprise positions of the P waves, the QRS waves and the T waves, or waveforms of the P waves, the QRS waves and the T waves are not complete.
The present invention uses the two-stage scheme (compared to the one-stage scheme, the precision is higher than that of the one-stage scheme, but the speed is slower), and the result obtained in step S2 is whether the pre-selected window contains the primary classification of one of the P wave, the QRS wave, and the T wave, and the detected waveforms are not distinguished. Thus, at level rpn, a classifier _ position layer follows. Corresponding target waveforms are found on P1 and P2 through the suggestion box, the lengths of the intercepted target waveforms are unequal, the intercepted target waveform data need to be zoomed to the same length, then the detected target waveforms are classified, and the target waveforms are determined to be one of P waves, QRS waves and T waves. And calculating the part of the target window through the layer to obtain more accurate position information to obtain a prediction window.
Step S6: performing suggested frame offset dataset processing
There may be multiple instances of repetition or overlap of proposed box positions derived from the previous layer. In the deep learning target detection algorithm, a non-maximum suppression mode is used for clearing the type of suggestion boxes, but because the suggestion box processing layer is not the final result, the threshold value of the used non-maximum suppression is large, partial suggestion boxes are reserved for further processing, and it is noted that the suggestion boxes obtained by the layer have small overlap ratio with the real P wave, QRS wave and T wave positions, so the suggestion boxes are further processed.
The present invention uses the two-stage detection scheme, so the output result at rpn level will be compared with the positive and negative samples and background calculated in step S2, and the offset of the preselected window as the optimization target to calculate the loss value. Wherein the calculation of the window selection offset and the preliminary classification of whether the preselected window contains one of the waveforms P-wave, QRS-wave, and T-wave discards the background portion, focusing only on positive and negative samples. The preselected window offset and whether the preselected window contains P waves, QRS waves, T waves are obtained by rpn layers.
Step S7: suggesting a second treatment of the box
Before data enters a classification layer (classifier _ position layer), rpn layers of optimized results are reversely encoded by TargetStage, the results are converted into window widths from offsets, then nms non-maximum value suppression is carried out, and suggestion boxes with high overlapping rate but low scores are removed to prevent imbalance of positive and negative samples. For electrocardiographic signal data within 10 seconds, the maximum normal heart rate does not exceed 18 times, and a set of data is set to 100 targets, since the sum of the P-wave waveform data and the T-wave waveform data is about 54 times. Specifically, the rpn-layer output offset reverse coding is used for obtaining the intersection ratio calculation of the suggestion frame and the actually marked real frame, a window with the intersection ratio larger than 75% is reserved as a positive sample, the first 54 positive samples are screened out, and the 100 targets are filled with negative samples with the intersection ratio smaller than 54 and larger than 0% and smaller than 50%. Positive samples in these 100 targets are distinguished by comparison with the data of true markers, where positive samples belong to P-wave, QRS-wave, T-wave and are classified as 1, 2, 3 and negative samples are labeled as 0. And simultaneously, obtaining the recommended frame and the actual frame of the real mark through reverse coding, and calculating the offset again to be used as the optimization target data of the classifier _ position layer.
Step S8: and (4) inputting the output features acquired through the feature extraction network and the classified sample data in the step (S7) into a classification layer, and intercepting the extracted features through a secondarily processed suggestion box, wherein the length of the suggestion box is changed into the same size because the lengths of the suggestion boxes are different, and then the suggestion boxes are input into the classification layer, and the offset of the prediction box and the corresponding class of the prediction box are obtained through the classification layer.
Specifically, the feature layers P1, P2 in step S4 and the 100 target data reverse-encoded by rpn layer offset in step S7 are converted into window width data, and input into the classifier _ position layer. The length data is intercepted on different feature layers according to different window sizes in the 100 targets, and the intercepted feature data is used for calculating the loss value at the layer by calculating the offset again with the classification mark of the 100 targets and the actual frame of the real mark in the step S4 as the optimization target. By the step, a prediction window closer to the real mark and which class of P wave, QRS wave and T wave the target in the prediction window belongs to are obtained.
And finally obtaining the start and end positions and the types of the P wave, the QRS wave and the T wave through model prediction of the obtained electrocardiogram data.
Compared with the traditional method for detecting the P wave and the T wave, the invention has the following difference: the traditional method for detecting P waves and T waves relies on the use of a differential method to determine the QRS waves first, and if the QRS waves are not accurately detected, the detection of the P waves and the T waves is also influenced. After the QRS wave is detected, the P wave and the T wave are determined by calculating a difference threshold line and a vertex area threshold, but the P wave has small amplitude in the electrocardiosignal and is easy to interfere, so that the P wave is difficult to detect. In addition, the traditional method for detecting P waves, T waves and QRS waves adopts an empirical threshold mode, and the threshold setting adaptability is low. Data accuracy decreases when beyond the manually designed empirical threshold. However, deep learning techniques can improve the adaptivity of the algorithm through big data learning.
Compared with the deep learning electrocardiosignal detection algorithm which is disclosed at present, the invention has the following difference: most of the deep learning detection algorithms disclosed at present detect a single QRS wave or P wave, and the invention realizes that a single model simultaneously detects the P wave, the QRS wave and the T wave. Compared with a single deep learning P slice wave detection algorithm, the standard P wave time is 0.11s, but interference or special P waves can occur in the actual electrocardiosignals, so that the P wave time is greater than or less than 0.11s, therefore, different preselection windows for P wave detection are designed in the scheme, and P wave detection with different time lengths is met. In order to realize the simultaneous detection of P wave, QRS wave and T wave, the invention is different from the single wave detection deep learning algorithm which is disclosed at present. According to the P wave length is generally less than 0.11s, the QRS wave length is 0.11-0.25s, and the T wave length is about 0.25s, so that the P wave and the QRS wave and the T wave are detected by using different size characteristic layers, and different pre-selection windows for detecting the QRS wave and the T wave are designed.
It should be recognized that embodiments of the present invention can be realized and implemented by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable memory. The methods may be implemented in a computer program using standard programming techniques, including a non-transitory computer-readable storage medium configured with the computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner, according to the methods and figures described in the detailed description. Each program may be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on a programmed application specific integrated circuit for this purpose.
Further, the operations of processes described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and/or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) collectively executed on one or more processors, by hardware, or combinations thereof. The computer program includes a plurality of instructions executable by one or more processors.
Further, the method may be implemented in any type of computing platform operatively connected to a suitable interface, including but not limited to a personal computer, mini computer, mainframe, workstation, networked or distributed computing environment, separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the invention may be embodied in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optically read and/or write storage medium, RAM, ROM, or the like, such that it may be read by a programmable computer, which when read by the storage medium or device, is operative to configure and operate the computer to perform the procedures described herein. Further, the machine-readable code, or portions thereof, may be transmitted over a wired or wireless network. The invention described herein includes these and other different types of non-transitory computer-readable storage media when such media include instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. The invention also includes the computer itself when programmed according to the methods and techniques described herein.
A computer program can be applied to input data to perform the functions described herein to transform the input data to generate output data that is stored to non-volatile memory. The output information may also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including particular visual depictions of physical and tangible objects produced on a display.
Finally, the above embodiments are only intended to illustrate the technical solutions of the present invention and not to limit the present invention, and although the present invention has been described in detail with reference to the preferred embodiments, it will be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions, and all of them should be covered by the claims of the present invention.
Claims (10)
1. A method for simultaneously positioning the positions of electrocardio P waves, QRS waves and T waves based on deep learning multi-target detection is characterized by comprising the following steps of: the method comprises the following steps:
step S1: acquiring electrocardiosignals, and performing noise reduction processing on the electrocardiosignals;
step S2: intercepting the marked electrocardiogram data for a set time length, ensuring that an electrocardiogram sample set contains different conditions of P waves, QRS waves and T waves of all the electrocardiogram data, and marking the starting and ending positions of the P waves, the QRS waves and the T waves of the electrocardiogram data set to obtain a marked data set;
step S3: carrying out prior frame processing in deep learning target detection on the intercepted and marked sample data, wherein the width of the prior frame is related to the widths of a P wave, a QRS wave and a T wave, and a plurality of prior frames are generated on the sample data; converting the positions of the prior frame, the actual marked P wave, the QRS wave and the T wave into the relation of offset according to the deep learning target detection, thereby obtaining a data set required by model training;
step S4: using deep learning target detection to identify a feature extraction network used for the heartbeat detection;
step S5: inputting the training set data prepared in the step S3 into a model according to a suggestion frame detection algorithm in a deep learning target detection algorithm, and performing regression and classification operation on the model to distinguish which parts of the electrocardiogram data with set time length are suggestion frames containing P wave, QRS wave and T wave positions;
step S6: carrying out suggested frame offset data set processing;
step S7: performing secondary processing on the suggestion frames, performing intersection comparison calculation on the positions of the suggestion frames obtained in the last step and the positions of the P waves, the QRS waves and the T waves of the real marks, taking the suggestion frames which are larger than a certain threshold value as positive samples, marking the categories of the P waves, the QRS waves and the T waves of the positive samples, taking the suggestion frames which are smaller than the certain threshold value as negative samples, and coding the suggestion frames of the positive samples and the negative samples into a coefficient relation with the offset of the real frames to be used as sample data of a classification layer;
step S8: inputting the output features acquired through the feature extraction network and the classified sample data in the step S7 into a classification layer, intercepting the extracted features through a secondarily processed suggestion box, wherein the length of the suggestion box is changed into the same size and then is input into the classification layer because the lengths of the suggestion boxes are different, and obtaining the offset of the prediction box and the corresponding category of the prediction box through the classification layer;
step S9: and finally, obtaining the starting and ending positions and the types of the P wave, the QRS wave and the T wave by utilizing the electrocardiogram data through model prediction.
2. The method for positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection as claimed in claim 1, which is characterized in that: in step S4, the feature extraction network uses the ResNet structure.
3. The method for positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection as claimed in claim 1 or 2, wherein the method comprises the following steps: in step S4, an electrocardiogram data feature extraction network is implemented by using a pyramid-like network in which feature layers are superimposed one on top of another, two feature layers with different sizes are constructed, it is ensured that the set prior frame can contain the actually labeled P wave, QRS wave, and T wave positions, and the intercepted electrocardiogram data within the set time length is input to the feature extraction network to obtain electrocardiogram data features.
4. The method for positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection as claimed in claim 1, which is characterized in that: in step S6, according to the situation that there are multiple repetitions or overlaps in the position of the suggestion frame obtained in the previous layer, the non-maximum suppression mode is used in the deep learning target detection algorithm to clear the suggestion frame, but because the processing layer of the suggestion frame is not the final result, the threshold value of the used non-maximum suppression is large, and a part of the suggestion frame is retained for further processing.
5. The method for positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection as claimed in claim 1, which is characterized in that: in step S9, the prediction frames obtained may overlap, and these overlapping prediction frames are suppressed and eliminated by using the non-maximum value.
6. The method for positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection as claimed in claim 1, which is characterized in that: in the step S1, the set time length is 8-20S.
7. The method for simultaneously positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection as claimed in claim 1, wherein the method comprises the following steps: in the step S3, a plurality of preselected windows are obtained at the determined sampling frequency, the preselected windows and the marked data real windows are used to perform iou union ratio calculation in target detection, the union ratio calculation mode is a type of method in general image target detection, two-dimensional calculation is reduced to one-dimensional calculation in electrocardiographic data, union ratios are obtained through iou calculation, prediction windows in different union ratio ranges are divided into positive samples, negative samples and backgrounds are respectively marked, and preselected frames and real frames in the positive samples are calculated to obtain the offset of each preselected window to the real window.
8. The method for simultaneously positioning the positions of the P wave, the QRS wave and the T wave of the electrocardio based on the deep learning multi-target detection as claimed in claim 7, wherein the method comprises the following steps: the positive samples are labeled 1, the background is 0, and the negative samples are-1.
9. The utility model provides a device based on deep learning multi-target detection location electrocardio P ripples, QRS ripples, T ripples position, includes memory, treater and stores on the memory and can be at the computer program of treater operation which characterized in that: the processor, when executing the computer program, implements the method of any of claims 1-5.
10. A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program, when executed by a processor, implements the method of any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111662087.5A CN114176602B (en) | 2021-12-30 | Method for simultaneously positioning electrocardiograph P wave, QRS wave and T wave based on deep learning multi-target detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111662087.5A CN114176602B (en) | 2021-12-30 | Method for simultaneously positioning electrocardiograph P wave, QRS wave and T wave based on deep learning multi-target detection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114176602A true CN114176602A (en) | 2022-03-15 |
| CN114176602B CN114176602B (en) | 2024-04-26 |
Family
ID=
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115956925A (en) * | 2023-02-10 | 2023-04-14 | 合肥心之声健康科技有限公司 | QRS wave detection method and system based on multistage smooth envelope |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101919695A (en) * | 2010-08-06 | 2010-12-22 | 李楚雅 | Electrocardiosignal QRS complex detection method based on wavelet transform |
| JP2012210235A (en) * | 2011-03-30 | 2012-11-01 | Sony Corp | Signal processing device, signal processing method and program, and information processing device |
| CN103110417A (en) * | 2013-02-28 | 2013-05-22 | 华东师范大学 | Automatic electrocardiogram recognition system |
| US20150335288A1 (en) * | 2013-06-06 | 2015-11-26 | Tricord Holdings, Llc | Modular physiologic monitoring systems, kits, and methods |
| CN110013247A (en) * | 2019-05-24 | 2019-07-16 | 东北大学 | A kind of detection, differentiation and the localization method of P wave of electrocardiogram and T wave |
| CN111184508A (en) * | 2020-01-19 | 2020-05-22 | 武汉大学 | Electrocardiosignal detection device and analysis method based on joint neural network |
| CN111345808A (en) * | 2018-12-24 | 2020-06-30 | Zoll医疗公司 | Method for processing electrocardiosignal, electrocardiosignal monitoring device and storage medium |
| CN112826513A (en) * | 2021-01-05 | 2021-05-25 | 华中科技大学 | Fetal heart rate detection system based on deep learning and specificity correction on FECG |
| CN112842355A (en) * | 2021-02-24 | 2021-05-28 | 推演医疗科技(北京)有限责任公司 | Electrocardiosignal heart beat detection and identification method based on deep learning target detection |
| CN113128585A (en) * | 2021-04-16 | 2021-07-16 | 重庆康如来科技有限公司 | Deep neural network based multi-size convolution kernel method for realizing electrocardiographic abnormality detection and classification |
| CN113491523A (en) * | 2021-07-30 | 2021-10-12 | 济南汇医融工科技有限公司 | Electrocardiosignal characteristic point detection method and system |
| CN113499079A (en) * | 2021-06-18 | 2021-10-15 | 南京信息工程大学 | Atrial fibrillation detection method in electrocardiogram |
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101919695A (en) * | 2010-08-06 | 2010-12-22 | 李楚雅 | Electrocardiosignal QRS complex detection method based on wavelet transform |
| JP2012210235A (en) * | 2011-03-30 | 2012-11-01 | Sony Corp | Signal processing device, signal processing method and program, and information processing device |
| CN103110417A (en) * | 2013-02-28 | 2013-05-22 | 华东师范大学 | Automatic electrocardiogram recognition system |
| US20150335288A1 (en) * | 2013-06-06 | 2015-11-26 | Tricord Holdings, Llc | Modular physiologic monitoring systems, kits, and methods |
| CN111345808A (en) * | 2018-12-24 | 2020-06-30 | Zoll医疗公司 | Method for processing electrocardiosignal, electrocardiosignal monitoring device and storage medium |
| CN110013247A (en) * | 2019-05-24 | 2019-07-16 | 东北大学 | A kind of detection, differentiation and the localization method of P wave of electrocardiogram and T wave |
| CN111184508A (en) * | 2020-01-19 | 2020-05-22 | 武汉大学 | Electrocardiosignal detection device and analysis method based on joint neural network |
| CN112826513A (en) * | 2021-01-05 | 2021-05-25 | 华中科技大学 | Fetal heart rate detection system based on deep learning and specificity correction on FECG |
| CN112842355A (en) * | 2021-02-24 | 2021-05-28 | 推演医疗科技(北京)有限责任公司 | Electrocardiosignal heart beat detection and identification method based on deep learning target detection |
| CN113128585A (en) * | 2021-04-16 | 2021-07-16 | 重庆康如来科技有限公司 | Deep neural network based multi-size convolution kernel method for realizing electrocardiographic abnormality detection and classification |
| CN113499079A (en) * | 2021-06-18 | 2021-10-15 | 南京信息工程大学 | Atrial fibrillation detection method in electrocardiogram |
| CN113491523A (en) * | 2021-07-30 | 2021-10-12 | 济南汇医融工科技有限公司 | Electrocardiosignal characteristic point detection method and system |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115956925A (en) * | 2023-02-10 | 2023-04-14 | 合肥心之声健康科技有限公司 | QRS wave detection method and system based on multistage smooth envelope |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109117730B (en) | Real-time electrocardiogram atrial fibrillation judgment method, device and system and storage medium | |
| Wu et al. | On the adaptive detection of blood vessels in retinal images | |
| Zhu et al. | Detection of the optic disc in images of the retina using the Hough transform | |
| CN111340142B (en) | Epilepsia magnetoencephalogram spike automatic detection method and tracing positioning system | |
| CN109493343A (en) | Medical image abnormal area dividing method and equipment | |
| US11617528B2 (en) | Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN110680302B (en) | Automatic identification method for electrocardiosignal characteristic wave | |
| CN111481192B (en) | Electrocardiosignal R wave detection method based on improved U-Net | |
| CN110432895B (en) | Training data processing method, electrocardiographic waveform detection method and electronic equipment | |
| CN104182984B (en) | Method and system for rapidly and automatically collecting blood vessel edge forms in dynamic ultrasonic image | |
| CN111956208B (en) | ECG signal classification method based on ultra-lightweight convolutional neural network | |
| Guo et al. | Emfn: Enhanced multi-feature fusion network for hard exudate detection in fundus images | |
| CN113450305B (en) | Medical image processing method, system, equipment and readable storage medium | |
| Panhwar et al. | Plant health detection enabled CNN scheme in IoT network | |
| CN111899272B (en) | Fundus image blood vessel segmentation method based on coupling neural network and line connector | |
| CN103839048B (en) | Stomach CT image lymph gland recognition system and method based on low-rank decomposition | |
| CN114176602B (en) | Method for simultaneously positioning electrocardiograph P wave, QRS wave and T wave based on deep learning multi-target detection | |
| CN114176602A (en) | Method for simultaneously positioning positions of electrocardio P wave, QRS wave and T wave based on deep learning multi-target detection | |
| CN110916645A (en) | QRS wave identification method combining wavelet transformation and image segmentation network | |
| CN113128585B (en) | Deep neural network based multi-size convolution kernel method for realizing electrocardiographic abnormality detection and classification | |
| CN113647959B (en) | Waveform identification method, device and equipment for electrocardiographic waveform signals | |
| Ali et al. | Segmenting retinal blood vessels with gabor filter and automatic binarization | |
| CN108764311A (en) | A kind of shelter target detection method, electronic equipment, storage medium and system | |
| CN111640126B (en) | Artificial intelligent diagnosis auxiliary method based on medical image | |
| Ashame et al. | Abnormality Detection in Eye Fundus Retina |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |