WO2021143400A1 - 一种基于r点的心搏数据分类方法和装置 - Google Patents

一种基于r点的心搏数据分类方法和装置 Download PDF

Info

Publication number
WO2021143400A1
WO2021143400A1 PCT/CN2020/134748 CN2020134748W WO2021143400A1 WO 2021143400 A1 WO2021143400 A1 WO 2021143400A1 CN 2020134748 W CN2020134748 W CN 2020134748W WO 2021143400 A1 WO2021143400 A1 WO 2021143400A1
Authority
WO
WIPO (PCT)
Prior art keywords
heartbeat
recognition
index
identification
frame sequence
Prior art date
Application number
PCT/CN2020/134748
Other languages
English (en)
French (fr)
Inventor
张碧莹
田亮
曹君
Original Assignee
上海优加利健康管理有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 上海优加利健康管理有限公司 filed Critical 上海优加利健康管理有限公司
Priority to US17/758,944 priority Critical patent/US20220401005A1/en
Priority to EP20914272.8A priority patent/EP4091549A4/en
Publication of WO2021143400A1 publication Critical patent/WO2021143400A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—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
    • 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
    • A61B5/352—Detecting R peaks, e.g. for synchronising diagnostic apparatus; Estimating R-R interval
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • 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]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching
    • G06F2218/16—Classification; Matching by matching signal segments

Definitions

  • the present invention relates to the technical field of electrocardiographic signal processing, in particular to a method and device for classification of heartbeat data based on R point.
  • ECG data is a set of electrical signal data related to the cardiac cycle of the heart
  • ECG analysis is the characteristic analysis of the collected ECG data.
  • Conventional ECG data waveform has 5 characteristic points, which are respectively P, Q, R, S, and T points.
  • the ECG data analysis method is to first identify and extract the feature point information, and then select the classification method for classification.
  • the current identification method for the feature points of ECG data is to have the identification requirements for the above five feature points.
  • the output analysis report is also prone to feature omission and feature deviation.
  • the purpose of the present invention is to provide a heartbeat data classification method and device based on the R point in view of the defects of the prior art.
  • the R point signal of the strongest signal among the 5 points is used as the heartbeat signal feature point, and the maximum number is retained.
  • the heartbeat signal data of the system solves the problem of losing heartbeat data in the conventional method; secondly, the heartbeat signal with the R point as the main feature point can be further classified and output.
  • the first aspect of the embodiments of the present invention provides a heartbeat data classification method based on R point, characterized in that the method includes:
  • the first recognition frame sequence includes a plurality of first recognition frames
  • the second recognition The frame sequence includes a plurality of the second identification frames;
  • the second identification frame includes an R heartbeat classification probability group;
  • the heartbeat classification probability group includes at least one type of heartbeat classification probability parameter;
  • the R point position information and effective parameter extraction processing is performed on all the second recognition frames of the second recognition frame sequence to generate the R point position and heartbeat data classification information sequence.
  • the one-dimensional ECG data whose acquisition time length is a preset segment time threshold generates an ECG data segment
  • a target detection algorithm is called to perform heartbeat signal data feature recognition processing on the ECG data segment to generate a first recognition Box sequence, specifically including:
  • the target detection algorithm and use the preset grid time threshold as the grid division step size to perform average grid division processing on the ECG data fragments to generate fragment grid groups, and perform heartbeat signal data feature recognition on the fragment grids
  • a plurality of said first identification frames are generated by processing, and all the first identification frames generated by all said fragment grids are counted according to the grid sequence to generate said first identification frame sequence
  • said fragment grid group includes a plurality of The fragment grid
  • the first identification frame includes the first heartbeat signal probability, R point relative time data, QRS normalized time width, and the first heartbeat classification probability group
  • the first identification frame sequence includes multiple The first identification frame.
  • said performing absolute value conversion processing on all the first recognition frames of the first recognition frame sequence to generate a second recognition frame sequence, and performing non-maximum value suppression processing on the second recognition frame sequence Specifically:
  • Step 31 Initialize the second identification frame sequence to be empty; initialize the value of the first index to 1, and initialize the first total to the total number of first identification frames in the first identification frame sequence;
  • Step 32 Set the first index second identification frame; initialize the second heartbeat signal probability of the first index second identification frame to be empty, and initialize the R point absolute time data of the first index second identification frame to be empty Initializing the QRS absolute time width of the first index second identification frame to be empty, and initializing the heartbeat classification probability group of the first index second identification frame to be empty;
  • Step 33 Set the second heartbeat signal probability of the second identification frame of the first index to the first heartbeat signal probability of the first identification frame corresponding to the first index of the first identification frame sequence; Setting the heartbeat classification probability group of the first index second identification frame to the first heartbeat classification probability group of the first identification frame corresponding to the first index of the first identification frame sequence;
  • Step 34 Extract the R point relative time data of the first identification frame corresponding to the first index of the first identification frame sequence to generate time offset data in the grid, and subtract 1 from the first index
  • the quotient divided by the preset unit grid identification frame number threshold is rounded to the result of the calculation plus 1 and the grid index to which the identification frame belongs is generated, and the difference of the grid index to which the identification frame belongs minus 1 is multiplied by the preset
  • the product of the grid time threshold is set to generate grid start time data
  • the R point absolute time data of the first index and second identification frame is set as the grid start time data plus the time in the grid Sum of offset data;
  • Step 35 Extract the QRS normalized time width of the first recognition frame corresponding to the first index of the first recognition frame sequence to generate a normalized time width of the time width, and set all values of the second recognition frame of the first index.
  • the QRS absolute time width is the product of the square of the normalized value of the time width multiplied by the preset segment time threshold;
  • Step 36 Add a recognition frame object to the second recognition frame sequence by adding the second recognition frame of the first index to the second recognition frame sequence;
  • Step 37 Add 1 to the value of the first index
  • Step 38 Determine whether the first index is greater than the first total, if the first index is greater than the first total, then go to step 39, if the first index is less than or equal to the first total, then Go to step 32;
  • Step 39 Perform sequential heartbeat signal probability polling on all the second identification frames of the second identification frame sequence, and the second heartbeat signal probability of the second identification frame currently polled exceeds When the heartbeat signal probability threshold range is preset, the second identification frame currently polled is deleted from the second identification frame sequence;
  • Step 40 Perform a pairwise comparison of all the second recognition frames of the second recognition frame sequence, and when the time overlap ratio of the two second recognition frames participating in the comparison exceeds a preset recognition frame overlap ratio threshold In the case of the range, the second identification frame with the lower probability of the second heartbeat signal of the two is deleted from the second identification frame sequence.
  • the marking processing of valid parameters and invalid parameters on all the heartbeat classification probability parameters of the heartbeat classification probability group of all the second identification frames of the second identification frame sequence specifically includes:
  • the R point position information and effective parameter extraction processing is performed on all the second recognition frames in the second recognition frame sequence in chronological order to generate the R point position and heartbeat data classification information sequence, specifically include:
  • Step 51 Re-order all the second recognition frames in the second recognition frame sequence in chronological order according to the R point absolute time data
  • Step 52 Initialize the position of the R point and the heartbeat data classification information sequence to be empty; initialize the first temporary sequence to be empty; initialize the value of the second index to 1, and initialize the second total to be the first of the second identification frame sequence 2. Total number of identification frames;
  • Step 53 Set the second index R point position and the heartbeat data classification information; initialize the second index R point position and the R point position information of the heartbeat data classification information to be empty, initialize the second index R point position and The QRS width information of the heartbeat data classification information is empty; the initial position of the second index R point and the effective heartbeat classification probability group of the heartbeat data classification information are empty;
  • Step 54 Count the total number of heartbeat classification probability parameters marked as the effective parameter in the heartbeat classification probability group of the second identification frame corresponding to the second index of the second identification frame sequence to generate the total number of effective classification parameters;
  • Step 55 Determine whether the total number of effective classification parameters is equal to 0, if the total number of effective classification parameters is greater than 0, go to step 56, and if the total number of effective classification parameters is equal to 0, go to step 58;
  • Step 56 Set the R point position of the second index R point position and the heartbeat data classification information as the R point of the second recognition frame corresponding to the second index of the second recognition frame sequence Absolute time data; set the QRS width information of the second index R point position and heartbeat data classification information as the QRS absolute of the second identification frame corresponding to the second index of the second identification frame sequence Time width; extract all the heartbeat classification probability parameters marked as the effective parameter in the heartbeat classification probability group of the second identification frame corresponding to the second index of the second identification frame sequence, and then to all Performing a heartbeat classification probability parameter addition operation on the effective heartbeat classification probability group of the second index R point position and the heartbeat data classification information;
  • Step 57 Add the R point position and heartbeat data classification information of the second index to the first temporary sequence
  • Step 58 adding 1 to the value of the second index
  • Step 59 Determine whether the second index is greater than the second total, if the second index is greater than the second total, then go to step 60, if the second index is less than or equal to the second total, then Go to step 53;
  • Step 60 Extract all the R point positions and heartbeat data classification information of the first temporary sequence, and sequentially add R point positions and heartbeat data classification information to the R point position and heartbeat data classification information sequence operate.
  • the first aspect of the embodiments of the present invention provides a heartbeat data classification method based on R points, which uses a target detection algorithm to identify and output heartbeat signal characteristic points (R points) in a fixed-length ECG data segment.
  • a sequence of identification boxes containing R point identification information After obtaining the recognition output result, the embodiment of the present invention further performs absolute value conversion processing and non-maximum value suppression processing on the recognition frame sequence to obtain an optimized recognition frame sequence; furthermore, in the optimized recognition frame sequence Mark the validity of the heartbeat classification parameters.
  • taking the R point in the heartbeat signal as the characteristic point of the heartbeat signal can retain the maximum number of heartbeat signal data, and the upper application can make a variety of analysis report output based on the position of the R point and the classification information sequence of the heartbeat data set up.
  • a second aspect of the embodiments of the present invention provides a device, the device including a memory and a processor, the memory is used to store a program, and the processor is used to execute the first aspect and the methods in each implementation manner of the first aspect.
  • a third aspect of the embodiments of the present invention provides a computer program product containing instructions, which when the computer program product runs on a computer, causes the computer to execute the first aspect and the methods in the implementation manners of the first aspect.
  • a fourth aspect of the embodiments of the present invention provides a computer-readable storage medium on which a computer program is stored.
  • the computer program implements the first aspect and the methods in the first aspect when executed by a processor. .
  • FIG. 1 is a schematic diagram of a heartbeat data classification method based on R point according to Embodiment 1 of the present invention
  • FIG. 2 is a schematic diagram of a heartbeat signal provided by an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of marking effective and invalid parameters for all the heartbeat classification probability parameters of the heartbeat classification probability group according to the second embodiment of the present invention.
  • FIG. 4 is a schematic structural diagram of a heartbeat data classification device based on R point according to Embodiment 3 of the present invention.
  • Fig. 1 is a schematic diagram of a heartbeat data classification method based on R point provided in the first embodiment of the present invention. The method mainly includes the following steps:
  • Step 1 Obtain one-dimensional ECG data whose time length is a preset segment time threshold to generate an ECG data segment, and call a target detection algorithm to perform heartbeat signal data feature recognition processing on the ECG data segment to generate a first identification frame sequence;
  • the first recognition frame sequence includes a plurality of first recognition frames
  • step 11 obtaining one-dimensional ECG data whose time length is a preset segment time threshold to generate an ECG data segment;
  • the one-dimensional ECG data mentioned in the article is generated by extracting the heartbeat signal time information in the ECG lead data, which is a piece of heartbeat signal data information whose length is a preset segment time threshold; as shown in Figure 2
  • a piece of heartbeat signal data is composed of multiple heartbeat signal data, and each heartbeat signal data includes 5 characteristic points P, Q, R, S, T, which are composed of The figure also shows that among these 5 points, the peak value of point R is the highest. Compared with points P and T, the anti-interference ability of point R is the strongest. Therefore, the recognition accuracy of the effective signal will be improved compared with the traditional five-point beat signal recognition method through the R point.
  • a set of R point absolute time data, QRS absolute time width and heartbeat classification probability group related to R point can be identified by calling the target detection algorithm first, and then the validity mark for the heartbeat classification probability group , And finally extract the R point information with the effective heartbeat classification probability group to form the R point position and the heartbeat data classification information sequence;
  • Step 12 Invoke the target detection algorithm, use the preset grid time threshold as the grid division step size to perform average grid division processing on the ECG data fragments to generate fragment grid groups, and perform heartbeat signal data feature recognition processing on the fragment grids Generate a plurality of first recognition frames, and count all the first recognition frames generated from all the fragment grids according to the grid sequence to generate a first recognition frame sequence;
  • the segment grid group includes a plurality of segment grids;
  • the first identification frame includes the first heartbeat signal probability, R point relative time data, QRS normalization time width, and the first heartbeat classification probability group;
  • the first identification frame sequence It includes a plurality of first identification frames.
  • the target detection algorithm involved in the embodiment of the present invention uses a heartbeat signal prediction network model, which is based on the convolutional neural network (Convolutional Neural Network, CNN) principle training implementation of.
  • the prediction network model divides the ECG data fragments of a fixed time length into several time grids, and predicts the R point of the ECG data in each time grid, and finally predicts several times in each time grid.
  • the identification frame here not only has its own time width information, but also includes the probability of the heartbeat signal (the probability that the ECG data in the current identification frame belongs to the heartbeat signal), R point relative time data (the relative displacement of the R point signal relative to the start time of the time grid in the current identification frame time period), QRS normalized time width (the R wave signal time width in the current identification frame time period is relative to the ECG data The normalized value of the segment time length, where the QRS time width indicates the R wave signal time width) and the heartbeat classification probability group (the possible probability of the ECG data in the current identification frame time period for multiple heartbeat classification); in addition,
  • the data input of the prediction network model is limited by the software and hardware resources, and the length of the data input is limited.
  • the principle of fragmentation processing is: specify the preset fragment time threshold according to the maximum input length of the predicted network model, obtain fixed-length one-dimensional ECG data according to the fragment time threshold to generate ECG data fragments, and then input the ECG data fragments
  • the R point information is predicted in the predictive network model.
  • the target detection algorithm uses the prediction network model to divide the ECG data fragments whose length is the preset fragment time threshold into several fragment grids, and the time length of each fragment grid is equal to the preset grid time threshold; prediction After the network model is rasterized, the ECG data in each segment grid is calculated for R-point prediction, and finally multiple R-point prediction recognition frames (first recognition frames) are predicted; as above, each prediction recognition
  • the box includes at least four data items: the probability of the first heartbeat signal, the relative time data of the R point, the QRS normalization time width and the first heartbeat classification probability group; finally, the target detection algorithm will all the grids Performing statistics on the predicted recognition frames of the grid generates the first recognition frame sequence;
  • the first recognition frame sequence includes Y ⁇ Z first recognition frames .
  • Step 2 Perform absolute value conversion processing on all first recognition frames of the first recognition frame sequence to generate a second recognition frame sequence, and perform non-maximum value suppression processing on the second recognition frame sequence;
  • the second identification frame sequence includes a plurality of second identification frames; the second identification frame includes an R dimbeat classification probability group; the heartbeat classification probability group includes at least one type of heartbeat classification probability parameter;
  • step 21 initializing the second identification frame sequence to be empty; initializing the value of the first index to 1, and initializing the first total to the total number of first identification frames of the first identification frame sequence;
  • Step 22 Set the first index second identification frame; initialize the probability of the second heartbeat signal of the first index second identification frame to be empty, initialize the R point absolute time data of the first index second identification frame to be empty, initialize the first index
  • the QRS absolute time width of the index second recognition frame is empty, and the heartbeat classification probability group of the first index second recognition frame is initialized to be empty;
  • Step 23 Set the probability of the second heartbeat signal of the second identification frame of the first index to the probability of the first heartbeat signal of the first identification frame corresponding to the first index of the first identification frame sequence; set the second identification frame of the first index
  • the heartbeat classification probability group of is the first heartbeat classification probability group of the first identification frame corresponding to the first index of the first identification frame sequence;
  • Step 24 Extract the R point relative time data of the first recognition frame corresponding to the first index of the first recognition frame sequence to generate the time offset data in the grid, and divide the difference of the first index minus 1 by the preset unit grid recognition The quotient of the frame number threshold is rounded to the result of the calculation plus 1 to generate the grid index to which the identification frame belongs, and the grid start is generated according to the product of the grid index of the identification frame minus 1 multiplied by the preset grid time threshold Time data, set the R point absolute time data of the first index and second identification frame as the sum of the grid start time data plus the time offset data in the grid;
  • the grid start time data (the grid index to which the identification frame belongs-1) * preset grid time threshold
  • R point absolute time data R point relative time data + grid start time data
  • the first identification frame sequence is to extract the identification frame composition of each grid in sequence according to the grid order
  • the grid index described in the current recognition frame needs to be determined, that is, the grid index to which the recognition frame belongs.
  • the grid index to which the recognition frame belongs
  • +1 1, indicating that the grid belonging to the identification frame 11 and the
  • Step 25 Extract the QRS normalized time width of the first recognition frame corresponding to the first index of the first recognition frame sequence to generate a normalized time width value, and set the QRS absolute time width of the second recognition frame of the first index to the normalized time width The square of the value multiplied by the product of the preset segment time threshold;
  • Step 26 Add the recognition frame object to the second recognition frame sequence by adding the second recognition frame of the first index to the second recognition frame sequence;
  • Step 27 Add 1 to the value of the first index
  • Step 28 Judge whether the first index is greater than the first total, if the first index is greater than the first total, go to step 29, and if the first index is less than or equal to the first total, go to step 22;
  • Step 29 Perform sequential heartbeat signal probability polling on all second identification frames of the second identification frame sequence, and the second heartbeat signal probability of the second identification frame currently polled exceeds the preset heartbeat signal probability threshold range When, delete the second identification frame currently polled from the second identification frame sequence;
  • Step 30 Perform a pairwise comparison of all the second recognition frames of the second recognition frame sequence.
  • the time overlap ratio of the two second recognition frames participating in the comparison exceeds the preset recognition frame overlap ratio threshold range, the two The second identification frame with a lower probability of the second heartbeat signal is deleted from the second identification frame sequence.
  • steps 21-30 are detailed explanations of step 2;
  • step 2 because the floating point calculation amount of convolution calculation is very large, in order to improve the calculation efficiency, the target detection algorithm uses the predictive network model to calculate the predictive recognition frame.
  • the time value of the R point and the QRS time width adopt a relative calculation method; therefore, if the absolute position information of the R point is to be extracted, the relative time data of the R point and the QRS in the first recognition frame in the first recognition frame sequence are required.
  • the normalized time width is converted into absolute values, and the converted recognition frame is defined as the second recognition frame; if the first recognition frame sequence includes Y ⁇ Z first recognition frames, then the second recognition frame sequence at this time also includes Y ⁇ Z second identification frames; the data structure of the second identification frame and the data structure of the first identification frame are summarized in the following table;
  • the second recognition frame sequence is generated, and step 2 continues the non-maximum value suppression processing on the second recognition frame sequence.
  • the processing process includes two optimization steps: 1. Perform the heartbeat signal probability optimization processing on the identification frame sequence, that is, determine the second identification frame whose second heartbeat signal probability does not meet the preset heartbeat signal probability threshold range as the optimization object, and remove it from the second identification frame sequence; 2 , The second identification frame sequence is further optimized for coincidence, that is, the second identification frame in the second identification frame sequence is compared in pairs.
  • the coincidence ratio is calculated, and the time overlap ratio exceeds
  • the second recognition frame with the lower probability of the second heartbeat signal of the two is judged as the optimized object, and it is removed from the second recognition frame sequence; suppose that the above is completed at this time
  • the second recognition frame sequence of the 2 optimization steps includes N second recognition frames, then N ⁇ Y ⁇ Z.
  • Step 3 Perform valid parameter and invalid parameter labeling processing on all heartbeat classification probability parameters of the heartbeat classification probability group of all second identification frames in the second identification frame sequence;
  • step 3 is to perform a further heartbeat type labeling operation for all second identification frames in the second identification frame sequence based on their heartbeat classification probability groups.
  • the heartbeat classification probability group of each second identification frame includes 4 heartbeat classification probability parameters: heartbeat classification A (B/C/D) probability parameter; in the second identification frame sequence for the second identification frame During polling, the probability parameters of heartbeat classification A (B/C/D) are compared numerically, and the largest value among the four is marked as valid parameters, and the rest are marked as invalid parameters.
  • Step 4 Perform R point position information and effective parameter extraction processing on all second recognition frames in the second recognition frame sequence in chronological order to generate R point position and heartbeat data classification information sequence;
  • step 41 according to the absolute time data of point R, reorder all the second recognition frames in the second recognition frame sequence in chronological order;
  • Step 42 Initialize the R point position and the heartbeat data classification information sequence to be empty; initialize the first temporary sequence to be empty; initialize the second index to be 1, and initialize the second total to be the total number of second recognition frames in the second recognition frame sequence ;
  • Step 43 Set the second index R point position and heartbeat data classification information; initialize the second index R point position and heartbeat data classification information to empty the R point position information, initialize the second index R point position and heartbeat data classification information
  • the QRS width information of the information is empty; the effective heartbeat classification probability group of the initial second index R point position and the heartbeat data classification information is empty;
  • Step 44 Count the total number of heartbeat classification probability parameters marked as effective parameters in the heartbeat classification probability group of the second identification frame corresponding to the second index of the second identification frame sequence, and generate the total number of effective classification parameters;
  • Step 45 Judge whether the total number of effective classification parameters is equal to 0, if the total number of effective classification parameters is greater than 0, go to step 46, and if the total number of effective classification parameters is equal to 0, go to step 48;
  • Step 46 Set the R point position of the second index and the R point position information of the heartbeat data classification information as the R point absolute time data of the second recognition frame corresponding to the second index of the second recognition frame sequence; set the second index R point
  • the QRS width information of the position and heartbeat data classification information is the QRS absolute time width of the second recognition frame corresponding to the second index of the second recognition frame sequence; the second recognition frame corresponding to the second index of the second recognition frame sequence is extracted All heartbeat classification probability parameters marked as effective parameters in the heartbeat classification probability group are sequentially added to the effective heartbeat classification probability group of the second index R point position and heartbeat data classification information;
  • Step 47 Add the R point position and heartbeat data classification information of the second index to the first temporary sequence
  • Step 48 Add 1 to the value of the second index
  • Step 49 Determine whether the second index is greater than the second total, if the second index is greater than the second total, go to step 50, and if the second index is less than or equal to the second total, go to step 43;
  • Step 50 Extract all R point positions and heartbeat data classification information of the first temporary sequence, and sequentially perform an R point position and heartbeat data classification information addition operation to the R point position and heartbeat data classification information sequence.
  • steps 41-50 are detailed explanations of the generation process of the R point position and heartbeat data classification information sequence, because the second identification frame sequence undergoes absolute value conversion and two optimization processes in step 2, where the second identification The frame order may not meet the chronological order, so first, all the second recognition frame data items in the second recognition frame sequence are re-ordered according to the R point absolute time data of the second recognition frame; then, the sorting is performed again After the completion of the second recognition frame sequence, perform the sequential extraction operation of the second recognition frame, and the extracted content is limited to the R point absolute time data of the second recognition frame, the QRS absolute time width parameter and the heartbeat classification probability group; Check whether there are valid heartbeat classification parameters in the beat classification probability group.
  • Fig. 3 is a schematic diagram of marking effective and invalid parameters for all heartbeat classification probability parameters of the heartbeat classification probability group provided in the second embodiment of the present invention.
  • the second embodiment is used to perform effective parameter and invalid parameter labeling on the heartbeat classification probability parameters.
  • the method steps of invalid parameter mark processing are further explained in detail. The method mainly includes the following steps:
  • Step 101 Obtain a second recognition frame sequence
  • the second identification frame sequence includes a plurality of second identification frames; the second identification frame includes a heartbeat classification probability group, and the heartbeat classification probability parameter includes at least one type of heartbeat classification probability parameter.
  • the heartbeat classification probability group of each second recognition frame includes 4 types of heartbeat classification probabilities Parameters, as shown in the following table:
  • Step 102 Initialize the value of the third index to 1, and initialize the third total to the total number of second recognition frames in the second recognition frame sequence.
  • Step 103 Perform maximum polling on the heartbeat classification probability parameter of the heartbeat classification probability group in the second identification frame of the third index, mark the heartbeat classification probability parameter with the largest value as a valid parameter, and set the value not as the maximum value.
  • the heartbeat classification probability parameter is marked as an invalid parameter.
  • step 103 is the process of specifically marking the heartbeat classification probability parameters, taking the third index as 1 as an example; as shown in Table 2, the first and second identification frame includes 4 heartbeat classification probability parameters, which are : The probability parameter of category 1 is 9%, the probability parameter of category 2 is 11%, the probability parameter of category 3 is 15%, and the probability parameter of category 4 is 65%; then the heartbeat classification probability group of the first and second identification frame has the largest value It is the category 4 probability parameter; then the category 4 probability parameter is further marked as valid parameters, and the remaining parameters are marked as invalid parameters; therefore, the marking results of the first and second identification boxes are shown in the following table;
  • Step 104 Add 1 to the value of the third index.
  • Step 105 Determine whether the third index is greater than the third total. If the third index is greater than the third total, go to step 106; if the third index is less than or equal to the third total, go to step 103.
  • Step 106 Output the marked second identification frame sequence to the sequence processing flow of the R point position of the ECG data segment.
  • FIG. 4 is a schematic structural diagram of a heartbeat data classification device based on R points provided in the third embodiment of the present invention.
  • the device includes a processor and a memory.
  • the memory can be connected to the processor through a bus.
  • the memory can be a non-volatile memory, such as a hard disk drive and a flash memory, in which software programs and device drivers are stored.
  • the software program can execute various functions of the foregoing method provided by the embodiments of the present invention; the device driver may be a network and interface driver.
  • the processor is used to execute a software program, and when the software program is executed, the method provided in the embodiment of the present invention can be implemented.
  • the embodiment of the present invention also provides a computer-readable storage medium.
  • a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the method provided in the embodiment of the present invention can be implemented.
  • the embodiment of the present invention also provides a computer program product containing instructions.
  • the processor is caused to execute the above method.
  • the embodiment of the present invention provides a method and device for classification of heartbeat data based on R point, which uses a target detection algorithm to perform target recognition on a heartbeat signal feature point (R point) in a fixed-length ECG data segment and outputs The identification frame sequence of the R point identification information. After obtaining the recognition output result, the embodiment of the present invention further performs absolute value conversion processing and non-maximum value suppression processing on the recognition frame sequence to obtain an optimized recognition frame sequence; furthermore, in the optimized recognition frame sequence Mark the validity of the heartbeat classification parameters. Finally, extract the position information of the characteristic point (R point) of the heartbeat signal and the effective heartbeat classification information from the marked identification frame sequence to generate the R point position and the heartbeat data classification information sequence.
  • taking the R point in the heartbeat signal as the characteristic point of the heartbeat signal can retain the maximum number of heartbeat signal data, and the upper application can make a variety of analysis report output based on the position of the R point and the classification information sequence of the heartbeat data set up.
  • the steps of the method or algorithm described in combination with the embodiments disclosed in this document can be implemented by hardware, a software module executed by a processor, or a combination of the two.
  • the software module can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or all areas in the technical field. Any other known storage media.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Cardiology (AREA)
  • Artificial Intelligence (AREA)
  • Veterinary Medicine (AREA)
  • Surgery (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Medical Informatics (AREA)
  • Molecular Biology (AREA)
  • Signal Processing (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Physics (AREA)
  • Physiology (AREA)
  • Psychiatry (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
  • Image Analysis (AREA)

Abstract

一种基于R点的心搏数据分类方法和装置，所述方法包括：获取时间长度为预置片段时间阈值的一维心电数据生成心电数据片段，并调用目标检测算法对心电数据片段进行心搏信号数据特征识别处理生成第一识别框序列；其中，第一识别框序列包括多个第一识别框（1）；对第一识别框序列的所有第一识别框进行绝对数值转换处理生成第二识别框序列，并对第二识别框序列进行非极大值抑制处理；其中，第二识别框序列包括多个第二识别框；第二识别框包括R点心搏分类概率组；心搏分类概率组包括至少一类心搏分类概率参数（2）；对第二识别框序列的所有第二识别框的心搏分类概率组的所有心搏分类概率参数进行有效参数与无效参数标记处理（3）；按时间先后顺序，对第二识别框序列的所有第二识别框进行R点位置信息及有效参数提取处理，生成R点位置及心搏数据分类信息序列（4）。

Description

一种基于R点的心搏数据分类方法和装置
本申请要求于2020年1月17日提交中国专利局、申请号为202010057333.3、发明名称为“一种基于R点的心搏数据分类方法和装置”的中国专利申请的优先权。
技术领域
本发明涉及心电信号处理技术领域，特别涉及一种基于R点的心搏数据分类方法和装置。
背景技术
心电数据是一组与心脏心动周期相关的电信号数据，心电分析是对采集的心电数据进行特征分析。常规心电数据波形有5个特征点，分别成为P、Q、R、S、T点。心电数据分析方式是先识别并提取特征点信息，然后再选择分类方法进行分类。目前针对心电数据特征点的识别方法是对上述5个特征点都有识别要求。在实际操作中，因为P点、T点信号容易受噪声信号干扰，在信号滤波降噪过程中被误消除的几率也较高，从而间接导致所属的心搏信号也会被丢失掉。在这种情况下，输出的分析报告也容易出现特征遗漏与特征偏差的问题。
发明内容
本发明的目的，就是针对现有技术的缺陷，提供一种基于R点的心搏数据分类方法和装置，首先以5点中最强信号R点信号作为心搏信号特征点，保留了最大数目的心搏信号数据作为分析源，解决了常规方法中丢失心搏数 据的问题；其次，对以R点为主要特征点的心搏信号还可以做进一步进行心搏分类输出。
为实现上述目的，本发明实施例第一方面提供了一种基于R点的心搏数据分类方法，其特征在于，所述方法包括：
获取时间长度为预置片段时间阈值的一维心电数据生成心电数据片段，并调用目标检测算法对所述心电数据片段进行心搏信号数据特征识别处理生成第一识别框序列；所述第一识别框序列包括多个第一识别框；
对所述第一识别框序列的所有所述第一识别框进行绝对数值转换处理生成第二识别框序列，并对所述第二识别框序列进行非极大值抑制处理；所述第二识别框序列包括多个所述第二识别框；所述第二识别框包括R点心搏分类概率组；所述心搏分类概率组包括至少一类心搏分类概率参数；
对所述第二识别框序列的所有所述第二识别框的所述心搏分类概率组的所有所述心搏分类概率参数进行有效参数与无效参数标记处理；
按时间先后顺序，对所述第二识别框序列的所有所述第二识别框进行R点位置信息及有效参数提取处理，生成R点位置及心搏数据分类信息序列。
优选的，所述获取时间长度为预置片段时间阈值的一维心电数据生成心电数据片段，并调用目标检测算法对所述心电数据片段进行心搏信号数据特征识别处理生成第一识别框序列，具体包括：
获取时间长度为所述预置片段时间阈值的所述一维心电数据生成所述心电数据片段；
调用所述目标检测算法，以预置栅格时间阈值为栅格划分步长对所述心电数据片段进行平均栅格划分处理生成片段栅格组，对片段栅格进行心搏信号数据特征识别处理生成多个所述第一识别框，按栅格先后顺序统计所有所述片段栅格生成的所有所述第一识别框生成所述第一识别框序列；所述片段栅格组包括多个所述片段栅格；所述第一识别框包括第一心搏信号概率、R点相对时间数据、QRS归一时间宽度和第一心搏分类概率组；所 述第一识别框序列包括多个所述第一识别框。
优选的，所述对所述第一识别框序列的所有所述第一识别框进行绝对数值转换处理生成第二识别框序列，并对所述第二识别框序列进行非极大值抑制处理，具体包括：
步骤31，初始化所述第二识别框序列为空；初始化第一索引的值为1，初始化第一总数为所述第一识别框序列的第一识别框总数；
步骤32，设置第一索引第二识别框；初始化所述第一索引第二识别框的第二心搏信号概率为空，初始化所述第一索引第二识别框的R点绝对时间数据为空，初始化所述第一索引第二识别框的QRS绝对时间宽度为空，初始化所述第一索引第二识别框的所述心搏分类概率组为空；
步骤33，设置所述第一索引第二识别框的所述第二心搏信号概率为所述第一识别框序列的第一索引对应的第一识别框的所述第一心搏信号概率；设置所述第一索引第二识别框的所述心搏分类概率组为所述第一识别框序列的所述第一索引对应的第一识别框的所述第一心搏分类概率组；
步骤34，提取所述第一识别框序列的所述第一索引对应的第一识别框的所述R点相对时间数据生成栅格内时间偏移数据，对所述第一索引减1的差除以预置单位栅格识别框数阈值的商做取整计算的结果加上1的和生成识别框所属栅格索引，根据所述识别框所属栅格索引减1的差乘以所述预置栅格时间阈值的乘积生成栅格起始时间数据，设置所述第一索引第二识别框的所述R点绝对时间数据为所述栅格起始时间数据加上所述栅格内时间偏移数据的和；
步骤35，提取所述第一识别框序列的所述第一索引对应的第一识别框的所述QRS归一时间宽度生成时间宽度归一值，设置所述第一索引第二识别框的所述QRS绝对时间宽度为所述时间宽度归一值的平方乘以所述预置片段时间阈值的乘积；
步骤36，将所述第一索引第二识别框向所述第二识别框序列进行识别 框对象添加操作；
步骤37，将所述第一索引的值加1；
步骤38，判断所述第一索引是否大于所述第一总数，如果所述第一索引大于所述第一总数则转至步骤39，如果所述第一索引小于或等于所述第一总数则转至步骤32；
步骤39，对所述第二识别框序列的所有所述第二识别框进行顺次心搏信号概率轮询，在当前轮询的所述第二识别框的所述第二心搏信号概率超出预置心搏信号概率阈值范围时，将当前轮询的所述第二识别框从所述第二识别框序列中删除；
步骤40，对所述第二识别框序列的所有所述第二识别框进行两两比对，当参与比对的两个所述第二识别框的时间重合比例超出预置识别框重合比例阈值范围时，将二者中所述第二心搏信号概率偏小的所述第二识别框从所述第二识别框序列中删除。
优选的，所述对所述第二识别框序列的所有所述第二识别框的所述心搏分类概率组的所有所述心搏分类概率参数进行有效参数与无效参数标记处理，具体包括：
对所述第二识别框序列的所有所述第二识别框进行依次轮询，将当前轮询的所述第二识别框的所述心搏分类概率组中数值最大的所述心搏分类概率参数标记为所述有效参数，当前轮询的所述第二识别框的所述心搏分类概率组中数值小于最大值的其他所述心搏分类概率参数标记为所述无效参数。
优选的，所述按时间先后顺序，对所述第二识别框序列的所有所述第二识别框进行R点位置信息及有效参数提取处理，生成R点位置及心搏数据分类信息序列，具体包括：
步骤51，根据所述R点绝对时间数据，按时间先后顺序对所述第二识别框序列中的所有所述第二识别框进行重新排序；
步骤52，初始化所述R点位置及心搏数据分类信息序列为空；初始化 第一临时序列为空；初始化第二索引的值为1，初始化第二总数为所述第二识别框序列的第二识别框总数；
步骤53，设置第二索引R点位置及心搏数据分类信息；初始化所述第二索引R点位置及心搏数据分类信息的R点位置信息为空，初始化所述第二索引R点位置及心搏数据分类信息的QRS宽度信息为空；初始化所述第二索引R点位置及心搏数据分类信息的有效心搏分类概率组为空；
步骤54，统计所述第二识别框序列的第二索引对应的第二识别框的所述心搏分类概率组中标记为所述有效参数的心搏分类概率参数总数，生成有效分类参数总数；
步骤55，判断所述有效分类参数总数是否等于0，如果所述有效分类参数总数大于0则转至步骤56，如果所述有效分类参数总数等于0则转至步骤58；
步骤56，设置所述第二索引R点位置及心搏数据分类信息的所述R点位置信息为所述第二识别框序列的所述第二索引对应的第二识别框的所述R点绝对时间数据；设置所述第二索引R点位置及心搏数据分类信息的所述QRS宽度信息为所述第二识别框序列的所述第二索引对应的第二识别框的所述QRS绝对时间宽度；提取所述第二识别框序列的所述第二索引对应的第二识别框的所述心搏分类概率组中标记为所述有效参数的所有心搏分类概率参数，顺次向所述第二索引R点位置及心搏数据分类信息的所述有效心搏分类概率组进行心搏分类概率参数添加操作；
步骤57，将所述第二索引R点位置及心搏数据分类信息向所述第一临时序列进行R点位置及心搏数据分类信息添加操作；
步骤58，将所述第二索引的值加1；
步骤59，判断所述第二索引是否大于所述第二总数，如果所述第二索引大于所述第二总数则转至步骤60，如果所述第二索引小于或等于所述第二总数则转至步骤53；
步骤60，提取所述第一临时序列的所有所述R点位置及心搏数据分类信息，顺次向所述R点位置及心搏数据分类信息序列进行R点位置及心搏数据分类信息添加操作。
本发明实施例第一方面提供的一种基于R点的心搏数据分类方法，利用目标检测算法对一个定长的心电数据片段中的心搏信号特征点(R点)进行目标识别并输出包含了R点识别信息的识别框序列。本发明实施例在获取了识别输出结果之后，进一步对识别框序列进行绝对数值转换处理和非极大值抑制处理从而获得优化后的识别框序列；再进一步的，在优化后的识别框序列中进行心搏分类参数有效性标记。最后，从标记后的识别框序列中提取心搏信号特征点(R点)位置信息与有效心搏分类信息生成R点位置及心搏数据分类信息序列。使用本发明方法，以心搏信号中的R点作为心搏信号特征点可以保留最大数目的心搏信号数据，上位应用进一步基于R点位置及心搏数据分类信息序列可做多种分析报告输出设置。
本发明实施例第二方面提供了一种设备，该设备包括存储器和处理器，存储器用于存储程序，处理器用于执行第一方面及第一方面的各实现方式中的方法。
本发明实施例第三方面提供了一种包含指令的计算机程序产品，当计算机程序产品在计算机上运行时，使得计算机执行第一方面及第一方面的各实现方式中的方法。
本发明实施例第四方面提供了一种计算机可读存储介质，计算机可读存储介质上存储有计算机程序，计算机程序被处理器执行时实现第一方面及第一方面的各实现方式中的方法。
附图说明
图1为本发明实施例一提供的一种基于R点的心搏数据分类方法示意图；
图2为本发明实施例提供的一种心搏信号示意图；
图3为本发明实施例二提供的对心搏分类概率组所有心搏分类概率参数进行有效参数与无效参数标记处理示意图；
图4为本发明实施例三提供的一种基于R点的心搏数据分类设备结构示意图。
具体实施方式
为了使本发明的目的、技术方案和优点更加清楚，下面将结合附图对本发明作进一步地详细描述，显然，所描述的实施例仅仅是本发明一部份实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
如图1为本发明实施例一提供的一种基于R点的心搏数据分类方法的示意图所示，本方法主要包括如下步骤：
步骤1，获取时间长度为预置片段时间阈值的一维心电数据生成心电数据片段，并调用目标检测算法对心电数据片段进行心搏信号数据特征识别处理生成第一识别框序列；
其中，第一识别框序列包括多个第一识别框；
具体包括：步骤11，获取时间长度为预置片段时间阈值的一维心电数据生成心电数据片段；
此处，文中提及的一维心电数据是通过提取心电导联数据中的心搏信号时间信息生成的，是一段长度为预置片段时间阈值的心搏信号数据信息；如图2为本发明实施例提供的一种心搏信号示意图所示，一段心搏信号数据由多个心搏信号数据组成，每个心搏信号数据包括5个特征点P、Q、R、S、T，由图也可见，这5点中，R点峰值最高，相较于P点与T点，R点的抗干扰能力是最强的。因此通过R点进行心搏信号识别比起传统的5点心搏信号识别方法来说，对有效信号的识别精度会提高。在本发明方法中， 首先通过调用目标检测算法可以识别出与R点相关的一组R点绝对时间数据、QRS绝对时间宽度和心搏分类概率组，然后对心搏分类概率组进行有效性标记，最终提取具备有效心搏分类概率组的R点信息组成R点位置及心搏数据分类信息序列；
步骤12，调用目标检测算法，以预置栅格时间阈值为栅格划分步长对心电数据片段进行平均栅格划分处理生成片段栅格组，对片段栅格进行心搏信号数据特征识别处理生成多个第一识别框，按栅格先后顺序统计所有片段栅格生成的所有第一识别框生成第一识别框序列；
其中，片段栅格组包括多个片段栅格；第一识别框包括第一心搏信号概率、R点相对时间数据、QRS归一时间宽度和第一心搏分类概率组；第一识别框序列包括多个第一识别框。
此处，对目标检测算法进行简要介绍：本发明实施例涉及的目标检测算法使用了一个心搏信号预测网络模型，该预测网络模型是基于卷积神经网络(Convolutional Neural Network,CNN)原理训练实现的。该预测网络模型通过将固定时间长度的心电数据片段平均分割成若干个时间栅格，对每个时间栅格内的心电数据进行R点预测，最终在每个时间栅格内预测出若干个R点预测识别框，理想的预测识别框内最多包含一个心搏信号也即最多包括一个R点。此处的识别框作为一个预测网络输出的数据对象，不仅仅具备自身的时间宽度信息，还包括了心搏信号概率(当前识别框时间段内的心电数据属于心搏信号的可能概率)、R点相对时间数据(当前识别框时间段内R点信号相对于时间栅格起始时间的相对位移)、QRS归一时间宽度(当前识别框时间段内R波信号时间宽度相对于心电数据片段时间长度的归一值，这里由QRS时间宽度标示R波信号时间宽度)和心搏分类概率组(当前识别框时间段内的心电数据针对多个心搏分类的可能概率)；另外，在具体实现中该预测网络模型的数据输入受软硬件资源限制是有长度限制的，如果一维心电数据的时间长度超过预测网络模型的输入 长度则需对一维心电数据进行片段化处理，片段化处理的原则是：根据预测网络模型输入长度最大值指定预置片段时间阈值，根据片段时间阈值获取定长的一维心电数据生成心电数据片段，再将心电数据片段划输入预测网络模型中进行R点信息预测。
此处，目标检测算法使用预测网络模型将长度为预置片段时间阈值的心电数据片段平均分割成若干个片段栅格，每个片段栅格的时间长度均等于预置栅格时间阈值；预测网络模型完成栅格化之后，再对每个片段栅格内的心电数据进行R点预测计算，最终预测出多个R点预测识别框(第一识别框)；如上文，每个预测识别框(第一识别框)都至少包括四个数据项：第一心搏信号概率、R点相对时间数据、QRS归一时间宽度和第一心搏分类概率组；最后，目标检测算法将所有栅格的预测识别框进行统计就生成了第一识别框序列；
假设，心电数据片段在目标检测算法实施过程中被划分为Y个栅格，每个栅格输出Z个第一识别框，则：第一识别框序列包括了Y×Z个第一识别框。
步骤2，对第一识别框序列的所有第一识别框进行绝对数值转换处理生成第二识别框序列，并对第二识别框序列进行非极大值抑制处理；
其中，第二识别框序列包括多个第二识别框；第二识别框包括R点心搏分类概率组；心搏分类概率组包括至少一类心搏分类概率参数；
具体包括：步骤21，初始化第二识别框序列为空；初始化第一索引的值为1，初始化第一总数为第一识别框序列的第一识别框总数；
步骤22，设置第一索引第二识别框；初始化第一索引第二识别框的第二心搏信号概率为空，初始化第一索引第二识别框的R点绝对时间数据为空，初始化第一索引第二识别框的QRS绝对时间宽度为空，初始化第一索引第二识别框的心搏分类概率组为空；
步骤23，设置第一索引第二识别框的第二心搏信号概率为第一识别框 序列的第一索引对应的第一识别框的第一心搏信号概率；设置第一索引第二识别框的心搏分类概率组为第一识别框序列的第一索引对应的第一识别框的第一心搏分类概率组；
步骤24，提取第一识别框序列的第一索引对应的第一识别框的R点相对时间数据生成栅格内时间偏移数据，对第一索引减1的差除以预置单位栅格识别框数阈值的商做取整计算的结果加上1的和生成识别框所属栅格索引，根据识别框所属栅格索引减1的差乘以预置栅格时间阈值的乘积生成栅格起始时间数据，设置第一索引第二识别框的R点绝对时间数据为栅格起始时间数据加上栅格内时间偏移数据的和；
此处，栅格起始时间数据＝(识别框所属栅格索引-1)*预置栅格时间阈值，R点绝对时间数据＝R点相对时间数据+栅格起始时间数据；
此处，因为一个栅格会输出多个识别框，具体数量以预置单位栅格识别框数阈值为准，第一识别框序列是按栅格顺序依次提取每个栅格里的识别框组成的识别框序列，在对每个识别框进行R点相对时间数据转换时，需要将当前识别框所述的栅格索引确定下来即识别框所属栅格索引，此处识别框所属栅格索引＝|(第一索引-1)/预置单位栅格识别框数阈值|+1；例如，一个心电数据片段包括Y个片段栅格，一个片段栅格包括Z个第一识别框，其中Z就是预置单位栅格识别框数阈值，第一识别框序列包括Y×Z个第一识别框，每连续Z个第一识别框分属一个片段栅格；又例如，一个心电数据片段包括3个片段栅格，一个片段栅格包括2个第一识别框，那么第一识别框序列包括3×2＝6个第一识别框{识别框 11，识别框 12，识别框 21，识别框 22，识别框 31，识别框 32}，当第一索引为1和2时，第一索引第一识别框分别为识别框 11和识别框 12，则识别框所属栅格索引＝|(1-1)/2|+1＝1和识别框所属栅格索引＝|(2-1)/2|+1＝1，表示识别框 11和识别框 12所述属的栅格是第1栅格；当第一索引为3和4时，识别框 13和识别框 14所述属的栅格是第2栅格；当第一索引为5和6时，识别框 15和 识别框 16所述属的栅格是第3栅格；
步骤25，提取第一识别框序列的第一索引对应的第一识别框的QRS归一时间宽度生成时间宽度归一值，设置第一索引第二识别框的QRS绝对时间宽度为时间宽度归一值的平方乘以预置片段时间阈值的乘积；
步骤26，将第一索引第二识别框向第二识别框序列进行识别框对象添加操作；
步骤27，将第一索引的值加1；
步骤28，判断第一索引是否大于第一总数，如果第一索引大于第一总数则转至步骤29，如果第一索引小于或等于第一总数则转至步骤22；
步骤29，对第二识别框序列的所有第二识别框进行顺次心搏信号概率轮询，在当前轮询的第二识别框的第二心搏信号概率超出预置心搏信号概率阈值范围时，将当前轮询的第二识别框从第二识别框序列中删除；
步骤30，对第二识别框序列的所有第二识别框进行两两比对，当参与比对的两个第二识别框的时间重合比例超出预置识别框重合比例阈值范围时，将二者中第二心搏信号概率偏小的第二识别框从第二识别框序列中删除。
此处，步骤21-30是对步骤2的详解；在步骤2中，因为卷积计算的浮点计算量非常庞大，为了提高计算效率，目标检测算法使用预测网络模型进行预测识别框计算时对R点的时间数值和QRS时间宽度采用了相对计算方式；由此，如果要提取R点的绝对位置信息就需要对第一识别框序列中的第一识别框中的R点相对时间数据和QRS归一时间宽度做绝对数值转换处理，转换后的识别框定义为第二识别框；假设第一识别框序列包括Y×Z个第一识别框的话，则此时的第二识别框序列也包括Y×Z个第二识别框；第二识别框的数据结构与第一识别框的数据结构简介如下表所示；
Figure PCTCN2020134748-appb-000001
Figure PCTCN2020134748-appb-000002
表一
在对第一识别框序列进行绝对值转换之后生成第二识别框序列，步骤2对第二识别框序列继续进行非极大值抑制处理，该处理过程包括两个优化步骤：1、对第二识别框序列进行心搏信号概率优化处理，即将第二心搏信号概率不满足预置心搏信号概率阈值范围的第二识别框判定为优化对象，将其从第二识别框序列中剔除；2、进一步对第二识别框序列进行重合优化处理，即对第二识别框序列中的第二识别框进行两两比对，如果二者时间段上有重合则计算重合比例，在时间重合比例超出预置识别框重合比 例阈值范围时，将二者中第二心搏信号概率偏小的那个第二识别框判定为优化对象，将其从第二识别框序列中剔除；假设此时，完成上述2个优化步骤的第二识别框序列包括N个第二识别框，则N≤Y×Z。
步骤3，对第二识别框序列的所有第二识别框的心搏分类概率组的所有心搏分类概率参数进行有效参数与无效参数标记处理；
具体包括：对第二识别框序列的所有第二识别框进行依次轮询，将当前轮询的第二识别框的心搏分类概率组中数值最大的心搏分类概率参数标记为有效参数，当前轮询的第二识别框的心搏分类概率组中数值小于最大值的其他心搏分类概率参数标记为无效参数。
此处，步骤3是对第二识别框序列中的所有第二识别框，基于其心搏分类概率组做进一步的心搏类型标记操作。例如，每个第二识别框的心搏分类概率组中包括4个心搏分类概率参数：心搏分类A(B/C/D)概率参数；在对第二识别框序列进行第二识别框轮询时，对心搏分类A(B/C/D)概率参数进行数值比对，将四者中数值最大的标记为有效参数，其余则标记为无效参数。
步骤4，按时间先后顺序，对第二识别框序列的所有第二识别框进行R点位置信息及有效参数提取处理，生成R点位置及心搏数据分类信息序列；
具体包括：步骤41，根据R点绝对时间数据，按时间先后顺序对第二识别框序列中的所有第二识别框进行重新排序；
步骤42，初始化R点位置及心搏数据分类信息序列为空；初始化第一临时序列为空；初始化第二索引的值为1，初始化第二总数为第二识别框序列的第二识别框总数；
步骤43，设置第二索引R点位置及心搏数据分类信息；初始化第二索引R点位置及心搏数据分类信息的R点位置信息为空，初始化第二索引R点位置及心搏数据分类信息的QRS宽度信息为空；初始化第二索引R点位置及心搏数据分类信息的有效心搏分类概率组为空；
步骤44，统计第二识别框序列的第二索引对应的第二识别框的心搏分类概率组中标记为有效参数的心搏分类概率参数总数，生成有效分类参数总数；
步骤45，判断有效分类参数总数是否等于0，如果有效分类参数总数大于0则转至步骤46，如果有效分类参数总数等于0则转至步骤48；
步骤46，设置第二索引R点位置及心搏数据分类信息的R点位置信息为第二识别框序列的第二索引对应的第二识别框的R点绝对时间数据；设置第二索引R点位置及心搏数据分类信息的QRS宽度信息为第二识别框序列的第二索引对应的第二识别框的QRS绝对时间宽度；提取第二识别框序列的第二索引对应的第二识别框的心搏分类概率组中标记为有效参数的所有心搏分类概率参数，顺次向第二索引R点位置及心搏数据分类信息的有效心搏分类概率组进行心搏分类概率参数添加操作；
步骤47，将第二索引R点位置及心搏数据分类信息向第一临时序列进行R点位置及心搏数据分类信息添加操作；
步骤48，将第二索引的值加1；
步骤49，判断第二索引是否大于第二总数，如果第二索引大于第二总数则转至步骤50，如果第二索引小于或等于第二总数则转至步骤43；
步骤50，提取第一临时序列的所有R点位置及心搏数据分类信息，顺次向R点位置及心搏数据分类信息序列进行R点位置及心搏数据分类信息添加操作。
此处，步骤41-50是对R点位置及心搏数据分类信息序列的生成过程详解，因为第二识别框序列在步骤2中经过绝对值转换和2次优化处理之后，其中的第二识别框顺序有可能不满足时间先后顺序，所以首先对第二识别框序列的中的所有第二识别框数据项根据第二识别框的R点绝对时间数据进行先后顺序重新排序；然后，再对排序完成之后的第二识别框序列进行第二识别框的依次提取操作，提取的内容仅限于第二识别框的R点绝 对时间数据、QRS绝对时间宽度参数和心搏分类概率组；接着，对心搏分类概率组中是否存在有效的心搏分类参数做检索，如果存在则提取将当前的第二识别框的心搏分类概率组中的有效心搏分类概率参数生成有效心搏分类概率组，并将当前第二识别框的R点绝对时间数据、QRS绝对时间宽度参数和有效心搏分类概率组向R点位置及心搏数据分类信息序列添加。
如图3为本发明实施例二提供的对心搏分类概率组所有心搏分类概率参数进行有效参数与无效参数标记处理示意图所示，实施例二用以对心搏分类概率参数进行有效参数与无效参数标记处理的方法步骤进行进一步详解说明，本方法主要包括如下步骤：
步骤101，获取第二识别框序列；
其中，第二识别框序列包括多个第二识别框；第二识别框包括心搏分类概率组，心搏分类概率参数至少包括一类心搏分类概率参数。
假设第二识别框序列包括4个第二识别框(识别框1，识别框2，识别框3，识别框4)，每个第二识别框的心搏分类概率组包括4类心搏分类概率参数，如下表所示：
Figure PCTCN2020134748-appb-000003
表二
步骤102，初始化第三索引的值为1，初始化第三总数为第二识别框序列的第二识别框总数。
步骤103，对第三索引第二识别框的心搏分类概率组的心搏分类概率参数进行最大值轮询，将数值最大的心搏分类概率参数标记为有效参数，将数值不为最大值的心搏分类概率参数标记为无效参数。
此处，步骤103是具体对心搏分类概率参数做标记的处理过程，以第三索引为1举例说明；如表二所示第1第二识别框包括4个心搏分类概率参数，分别是：分类1概率参数为9％，分类2概率参数为11％，分类3概率参数为15％，分类4概率参数为65％；则第1第二识别框的心搏分类概率组中数值最大的是分类4概率参数；则进一步将分类4概率参数标记为有效参数，其余参数标记为无效参数；由此，第1第二识别框的标记结果如下表所示；
Figure PCTCN2020134748-appb-000004
表三
步骤104，将第三索引的值加1。
步骤105，判断第三索引是否大于第三总数，如果第三索引大于第三总数转至步骤106，如果第三索引小于或等于第三总数转至步骤103。
步骤106，将完成标记的第二识别框序列输出至心电数据片段R点位置序列处理流程。
此处，通过步骤101-106的标记操作，针对表二代表的第二识别框序列的标记结果如下表所示：
Figure PCTCN2020134748-appb-000005
表四
图4为本发明实施例三提供的一种基于R点的心搏数据分类设备结构示意图，该设备包括：处理器和存储器。存储器可通过总线与处理器连接。存储器可以是非易失存储器，例如硬盘驱动器和闪存，存储器中存储有软 件程序和设备驱动程序。软件程序能够执行本发明实施例提供的上述方法的各种功能；设备驱动程序可以是网络和接口驱动程序。处理器用于执行软件程序，该软件程序被执行时，能够实现本发明实施例提供的方法。
需要说明的是，本发明实施例还提供了一种计算机可读存储介质。该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器执行时，能够实现本发明实施例提供的方法。
本发明实施例还提供了一种包含指令的计算机程序产品。当该计算机程序产品在计算机上运行时，使得处理器执行上述方法。
本发明实施例提供的一种基于R点的心搏数据分类方法和装置，利用目标检测算法对一个定长的心电数据片段中的心搏信号特征点(R点)进行目标识别并输出包含了R点识别信息的识别框序列。本发明实施例在获取了识别输出结果之后，进一步对识别框序列进行绝对数值转换处理和非极大值抑制处理从而获得优化后的识别框序列；再进一步的，在优化后的识别框序列中进行心搏分类参数有效性标记。最后，从标记后的识别框序列中提取心搏信号特征点(R点)位置信息与有效心搏分类信息生成R点位置及心搏数据分类信息序列。使用本发明方法，以心搏信号中的R点作为心搏信号特征点可以保留最大数目的心搏信号数据，上位应用进一步基于R点位置及心搏数据分类信息序列可做多种分析报告输出设置。
专业人员应该还可以进一步意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、计算机软件或者二者的结合来实现，为了清楚地说明硬件和软件的可互换性，在上述说明中已经按照功能一般性地描述了各示例的组成及步骤。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
结合本文中所公开的实施例描述的方法或算法的步骤可以用硬件、处 理器执行的软件模块，或者二者的结合来实施。软件模块可以置于随机存储器(RAM)、内存、只读存储器(ROM)、电可编程ROM、电可擦除可编程ROM、寄存器、硬盘、可移动磁盘、CD-ROM、或技术领域内所公知的任意其它形式的存储介质中。
以上所述的具体实施方式，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施方式而已，并不用于限定本发明的保护范围，凡在本发明的精神和原则之内，所做的任何修改、等同替换、改进等，均应包含在本发明的保护范围之内。

Claims (8)

  1. 一种基于R点的心搏数据分类方法，其特征在于，所述方法包括：
    获取时间长度为预置片段时间阈值的一维心电数据生成心电数据片段，并调用目标检测算法对所述心电数据片段进行心搏信号数据特征识别处理生成第一识别框序列；所述第一识别框序列包括多个第一识别框；
    对所述第一识别框序列的所有所述第一识别框进行绝对数值转换处理生成第二识别框序列，并对所述第二识别框序列进行非极大值抑制处理；所述第二识别框序列包括多个所述第二识别框；所述第二识别框包括R点心搏分类概率组；所述心搏分类概率组包括至少一类心搏分类概率参数；
    对所述第二识别框序列的所有所述第二识别框的所述心搏分类概率组的所有所述心搏分类概率参数进行有效参数与无效参数标记处理；
    按时间先后顺序，对所述第二识别框序列的所有所述第二识别框进行R点位置信息及有效参数提取处理，生成R点位置及心搏数据分类信息序列。
  2. 根据权利要求1所述的基于R点的心搏数据分类方法，其特征在于，所述获取时间长度为预置片段时间阈值的一维心电数据生成心电数据片段，并调用目标检测算法对所述心电数据片段进行心搏信号数据特征识别处理生成第一识别框序列，具体包括：
    获取时间长度为所述预置片段时间阈值的所述一维心电数据生成所述心电数据片段；
    调用所述目标检测算法，以预置栅格时间阈值为栅格划分步长对所述心电数据片段进行平均栅格划分处理生成片段栅格组，对片段栅格进行心搏信号数据特征识别处理生成多个所述第一识别框，按栅格先后顺序统计所有所述片段栅格生成的所有所述第一识别框生成所述第一识别框序列；所述片段栅格组包括多个所述片段栅格；所述第一识别框包括第一心搏信号概率、R点相对时间数据、QRS归一时间宽度和第一心搏分类概率组；所述第一识别框序列包括多个所述第一识别框。
  3. 根据权利要求2所述的基于R点的心搏数据分类方法，其特征在于，所述对所述第一识别框序列的所有所述第一识别框进行绝对数值转换处理生成第二识别框序列，并对所述第二识别框序列进行非极大值抑制处理，具体包括：
    步骤31，初始化所述第二识别框序列为空；初始化第一索引的值为1，初始化第一总数为所述第一识别框序列的第一识别框总数；
    步骤32，设置第一索引第二识别框；初始化所述第一索引第二识别框的第二心搏信号概率为空，初始化所述第一索引第二识别框的R点绝对时间数据为空，初始化所述第一索引第二识别框的QRS绝对时间宽度为空，初始化所述第一索引第二识别框的所述心搏分类概率组为空；
    步骤33，设置所述第一索引第二识别框的所述第二心搏信号概率为所述第一识别框序列的第一索引对应的第一识别框的所述第一心搏信号概率；设置所述第一索引第二识别框的所述心搏分类概率组为所述第一识别框序列的所述第一索引对应的第一识别框的所述第一心搏分类概率组；
    步骤34，提取所述第一识别框序列的所述第一索引对应的第一识别框的所述R点相对时间数据生成栅格内时间偏移数据，对所述第一索引减1的差除以预置单位栅格识别框数阈值的商做取整计算的结果加上1的和生成识别框所属栅格索引，根据所述识别框所属栅格索引减1的差乘以所述预置栅格时间阈值的乘积生成栅格起始时间数据，设置所述第一索引第二识别框的所述R点绝对时间数据为所述栅格起始时间数据加上所述栅格内时间偏移数据的和；
    步骤35，提取所述第一识别框序列的所述第一索引对应的第一识别框的所述QRS归一时间宽度生成时间宽度归一值，设置所述第一索引第二识别框的所述QRS绝对时间宽度为所述时间宽度归一值的平方乘以所述预置片段时间阈值的乘积；
    步骤36，将所述第一索引第二识别框向所述第二识别框序列进行识别 框对象添加操作；
    步骤37，将所述第一索引的值加1；
    步骤38，判断所述第一索引是否大于所述第一总数，如果所述第一索引大于所述第一总数则转至步骤39，如果所述第一索引小于或等于所述第一总数则转至步骤32；
    步骤39，对所述第二识别框序列的所有所述第二识别框进行顺次心搏信号概率轮询，在当前轮询的所述第二识别框的所述第二心搏信号概率超出预置心搏信号概率阈值范围时，将当前轮询的所述第二识别框从所述第二识别框序列中删除；
    步骤40，对所述第二识别框序列的所有所述第二识别框进行两两比对，当参与比对的两个所述第二识别框的时间重合比例超出预置识别框重合比例阈值范围时，将二者中所述第二心搏信号概率偏小的所述第二识别框从所述第二识别框序列中删除。
  4. 根据权利要求1所述的基于R点的心搏数据分类方法，其特征在于，所述对所述第二识别框序列的所有所述第二识别框的所述心搏分类概率组的所有所述心搏分类概率参数进行有效参数与无效参数标记处理，具体包括：
    对所述第二识别框序列的所有所述第二识别框进行依次轮询，将当前轮询的所述第二识别框的所述心搏分类概率组中数值最大的所述心搏分类概率参数标记为所述有效参数，当前轮询的所述第二识别框的所述心搏分类概率组中数值小于最大值的其他所述心搏分类概率参数标记为所述无效参数。
  5. 根据权利要求3所述的基于R点的心搏数据分类方法，其特征在于，所述按时间先后顺序，对所述第二识别框序列的所有所述第二识别框进行R点位置信息及有效参数提取处理，生成R点位置及心搏数据分类信息序列，具体包括：
    步骤51，根据所述R点绝对时间数据，按时间先后顺序对所述第二识 别框序列中的所有所述第二识别框进行重新排序；
    步骤52，初始化所述R点位置及心搏数据分类信息序列为空；初始化第一临时序列为空；初始化第二索引的值为1，初始化第二总数为所述第二识别框序列的第二识别框总数；
    步骤53，设置第二索引R点位置及心搏数据分类信息；初始化所述第二索引R点位置及心搏数据分类信息的R点位置信息为空，初始化所述第二索引R点位置及心搏数据分类信息的QRS宽度信息为空；初始化所述第二索引R点位置及心搏数据分类信息的有效心搏分类概率组为空；
    步骤54，统计所述第二识别框序列的第二索引对应的第二识别框的所述心搏分类概率组中标记为所述有效参数的心搏分类概率参数总数，生成有效分类参数总数；
    步骤55，判断所述有效分类参数总数是否等于0，如果所述有效分类参数总数大于0则转至步骤56，如果所述有效分类参数总数等于0则转至步骤58；
    步骤56，设置所述第二索引R点位置及心搏数据分类信息的所述R点位置信息为所述第二识别框序列的所述第二索引对应的第二识别框的所述R点绝对时间数据；设置所述第二索引R点位置及心搏数据分类信息的所述QRS宽度信息为所述第二识别框序列的所述第二索引对应的第二识别框的所述QRS绝对时间宽度；提取所述第二识别框序列的所述第二索引对应的第二识别框的所述心搏分类概率组中标记为所述有效参数的所有心搏分类概率参数，顺次向所述第二索引R点位置及心搏数据分类信息的所述有效心搏分类概率组进行心搏分类概率参数添加操作；
    步骤57，将所述第二索引R点位置及心搏数据分类信息向所述第一临时序列进行R点位置及心搏数据分类信息添加操作；
    步骤58，将所述第二索引的值加1；
    步骤59，判断所述第二索引是否大于所述第二总数，如果所述第二索 引大于所述第二总数则转至步骤60，如果所述第二索引小于或等于所述第二总数则转至步骤53；
    步骤60，提取所述第一临时序列的所有所述R点位置及心搏数据分类信息，顺次向所述R点位置及心搏数据分类信息序列进行R点位置及心搏数据分类信息添加操作。
  6. 一种设备，包括存储器和处理器，其特征在于，所述存储器用于存储程序，所述处理器用于执行如权利要求1至5任一项所述的方法。
  7. 一种包含指令的计算机程序产品，当其在计算机上运行时，使得计算机执行如权利要求1至5任一项所述的方法。
  8. 一种计算机可读存储介质，包括指令，当所述指令在计算机上运行时，使所述计算机执行根据权利要求1至5任一项所述的方法。
PCT/CN2020/134748 2020-01-17 2020-12-09 一种基于r点的心搏数据分类方法和装置 WO2021143400A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US17/758,944 US20220401005A1 (en) 2020-01-17 2020-12-09 Heartbeat data classification method and device based on point r
EP20914272.8A EP4091549A4 (en) 2020-01-17 2020-12-09 METHOD AND DEVICE FOR CLASSIFICATION OF HEARTBEAT DATA BASED ON A POINT R

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010057333.3 2020-01-17
CN202010057333.3A CN111259820B (zh) 2020-01-17 2020-01-17 一种基于r点的心搏数据分类方法和装置

Publications (1)

Publication Number Publication Date
WO2021143400A1 true WO2021143400A1 (zh) 2021-07-22

Family

ID=70948998

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/134748 WO2021143400A1 (zh) 2020-01-17 2020-12-09 一种基于r点的心搏数据分类方法和装置

Country Status (4)

Country Link
US (1) US20220401005A1 (zh)
EP (1) EP4091549A4 (zh)
CN (1) CN111259820B (zh)
WO (1) WO2021143400A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113647959A (zh) * 2021-07-27 2021-11-16 东软集团股份有限公司 一种心电波形信号的波形识别方法、装置及设备

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111259820B (zh) * 2020-01-17 2023-05-05 上海乐普云智科技股份有限公司 一种基于r点的心搏数据分类方法和装置
CN112244855B (zh) * 2020-11-13 2022-08-26 上海乐普云智科技股份有限公司 一种面向心电数据的数据处理系统
CN113892955B (zh) * 2021-10-18 2024-03-29 上海乐普云智科技股份有限公司 基于人工智能模型定位qrs波并进行分类的方法和装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107837082A (zh) * 2017-11-27 2018-03-27 乐普（北京）医疗器械股份有限公司 基于人工智能自学习的心电图自动分析方法和装置
CN109431487A (zh) * 2018-09-28 2019-03-08 上海优加利健康管理有限公司 心电图动态实时分析数据中室性心搏典型数据的识别方法
US20190082988A1 (en) * 2017-09-19 2019-03-21 Tata Consultancy Services Limited Cascaded binary classifier for identifying rhythms in a single-lead electrocardiogram (ecg) signal
CN110037690A (zh) * 2019-04-22 2019-07-23 上海数创医疗科技有限公司 基于改进卷积神经网络的r波定位方法和装置
CN110464333A (zh) * 2019-07-23 2019-11-19 深圳邦健生物医疗设备股份有限公司 一种心电图数据的存储方法及装置
CN111259820A (zh) * 2020-01-17 2020-06-09 上海优加利健康管理有限公司 一种基于r点的心搏数据分类方法和装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6115628A (en) * 1999-03-29 2000-09-05 Medtronic, Inc. Method and apparatus for filtering electrocardiogram (ECG) signals to remove bad cycle information and for use of physiologic signals determined from said filtered ECG signals
JP5920364B2 (ja) * 2012-02-01 2016-05-18 富士通株式会社 情報処理装置、代表波形生成方法および代表波形生成プログラム
US10194821B2 (en) * 2014-10-29 2019-02-05 Khalifa University of Science and Technology Medical device having automated ECG feature extraction
CN107822622B (zh) * 2017-09-22 2022-09-09 成都比特律动科技有限责任公司 基于深度卷积神经网络的心电图诊断方法和系统
CN108309262A (zh) * 2018-02-24 2018-07-24 乐普（北京）医疗器械股份有限公司 多参数监护数据分析方法和多参数监护仪

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190082988A1 (en) * 2017-09-19 2019-03-21 Tata Consultancy Services Limited Cascaded binary classifier for identifying rhythms in a single-lead electrocardiogram (ecg) signal
CN107837082A (zh) * 2017-11-27 2018-03-27 乐普（北京）医疗器械股份有限公司 基于人工智能自学习的心电图自动分析方法和装置
CN109431487A (zh) * 2018-09-28 2019-03-08 上海优加利健康管理有限公司 心电图动态实时分析数据中室性心搏典型数据的识别方法
CN110037690A (zh) * 2019-04-22 2019-07-23 上海数创医疗科技有限公司 基于改进卷积神经网络的r波定位方法和装置
CN110464333A (zh) * 2019-07-23 2019-11-19 深圳邦健生物医疗设备股份有限公司 一种心电图数据的存储方法及装置
CN111259820A (zh) * 2020-01-17 2020-06-09 上海优加利健康管理有限公司 一种基于r点的心搏数据分类方法和装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4091549A4

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113647959A (zh) * 2021-07-27 2021-11-16 东软集团股份有限公司 一种心电波形信号的波形识别方法、装置及设备
CN113647959B (zh) * 2021-07-27 2024-01-12 东软集团股份有限公司 一种心电波形信号的波形识别方法、装置及设备

Also Published As

Publication number Publication date
EP4091549A1 (en) 2022-11-23
US20220401005A1 (en) 2022-12-22
EP4091549A4 (en) 2024-03-20
CN111259820B (zh) 2023-05-05
CN111259820A (zh) 2020-06-09

Similar Documents

Publication Publication Date Title
WO2021143400A1 (zh) 一种基于r点的心搏数据分类方法和装置
Zhao et al. ECG feature extraction and classification using wavelet transform and support vector machines
EP4092570A1 (en) Heartbeat classification method for multi-tag ecg signal labeling, and device
EP3834736A1 (en) Method for determining cardiac cycles and ultrasonic device
CN103815897B (zh) 一种心电图特征提取方法
CN106108880B (zh) 一种心拍自动识别方法及系统
WO2023130662A1 (zh) 基于心脏超声视频筛选舒张期和收缩期图像的方法和装置
CN107133626B (zh) 一种基于部分平均随机优化模型的医学影像分类方法
CN113057648A (zh) 一种基于复合lstm结构的ecg信号分类方法
CN110688942A (zh) 一种基于InResNet网络的心电信号联合识别模块及方法
CN109124620A (zh) 一种房颤检测方法、装置及设备
CN112883803A (zh) 一种基于深度学习的心电信号分类方法、装置及存储介质
CN111291727A (zh) 一种光体积变化描记图法信号质量检测方法和装置
CN109887547B (zh) 一种基因序列比对滤波加速处理方法、系统及装置
CN108537123A (zh) 基于多特征提取的心电识别方法
CN113951893B (zh) 一种结合深度学习和电生理知识的多导联心电信号特征点提取方法
CN111134660B (zh) 一种r点位置的检测方法和装置
CN111078505B (zh) 一种监测数据处理方法、装置和计算设备
CN116725486A (zh) 一种基于双阶段的动态心电信号房颤检测方法、装置
CN111265211B (zh) 一种用于心搏数据的滑动检测方法和装置
CN111449647B (zh) 一种心电图识别方法、装置、存储介质及电子设备
CN106227990A (zh) 一种基于贪心策略的多粒度粗糙集简约方法及系统
CN112446307B (zh) 基于局部约束的非负矩阵分解的心电身份识别方法及系统
CN113647959A (zh) 一种心电波形信号的波形识别方法、装置及设备
CN113192629A (zh) 用于自动胎心判读的方法和装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20914272

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2020914272

Country of ref document: EP

Effective date: 20220817